BERT_Embeddings.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Second Experiment: BERT Embeddings.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "_Un-oDrHnDui",
        "r4_SDsRLasRA",
        "nZl09cJE0755",
        "X6uTwgD31Azk"
      ],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU",
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "6c05278600fe4b6da1ee0f25b22f5539": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_38a6d5d0831f49ed87a0188cfcbfce4c",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_a61785fe0a364969ab3d50cba4b9504d",
              "IPY_MODEL_099b7f85c72f4fddaceb5d2674485857"
            ]
          },
          "model_module_version": "1.5.0"
        },
        "38a6d5d0831f49ed87a0188cfcbfce4c": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          },
          "model_module_version": "1.2.0"
        },
        "a61785fe0a364969ab3d50cba4b9504d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_5b4c15cd098243338e685b0382abc91d",
            "_dom_classes": [],
            "description": "100%",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 405234788,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 405234788,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_7066203a1fac43f095ac16cc85d99e2d"
          },
          "model_module_version": "1.5.0"
        },
        "099b7f85c72f4fddaceb5d2674485857": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_21dab7daef1f40deb940bfc841eb3210",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 405M/405M [00:32&lt;00:00, 12.3MB/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_faa3b6da40364748b233e25ce23642bc"
          },
          "model_module_version": "1.5.0"
        },
        "5b4c15cd098243338e685b0382abc91d": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          },
          "model_module_version": "1.5.0"
        },
        "7066203a1fac43f095ac16cc85d99e2d": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          },
          "model_module_version": "1.2.0"
        },
        "21dab7daef1f40deb940bfc841eb3210": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          },
          "model_module_version": "1.5.0"
        },
        "faa3b6da40364748b233e25ce23642bc": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          },
          "model_module_version": "1.2.0"
        },
        "ac661179add94abfa964cec7fda930e9": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HBoxModel",
          "state": {
            "_view_name": "HBoxView",
            "_dom_classes": [],
            "_model_name": "HBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_dd5caeaee01a4dc0bb7fe2eb966405f9",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_e108dc18b08e424cb3a7c7c48f536dff",
              "IPY_MODEL_c8d3c5675def444bac6d8029abf11432"
            ]
          },
          "model_module_version": "1.5.0"
        },
        "dd5caeaee01a4dc0bb7fe2eb966405f9": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          },
          "model_module_version": "1.2.0"
        },
        "e108dc18b08e424cb3a7c7c48f536dff": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_829453c7bc7146fba2d76112fba0c5c9",
            "_dom_classes": [],
            "description": "",
            "_model_name": "FloatProgressModel",
            "bar_style": "success",
            "max": 1,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 1,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_3b3ee50161c247cb9489cb529d8d6dba"
          },
          "model_module_version": "1.5.0"
        },
        "c8d3c5675def444bac6d8029abf11432": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "HTMLModel",
          "state": {
            "_view_name": "HTMLView",
            "style": "IPY_MODEL_d5ded497a4f4490a8de285779dc5e501",
            "_dom_classes": [],
            "description": "",
            "_model_name": "HTMLModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 3/? [00:01&lt;00:00,  2.69it/s]",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_a4e7c79f80d4493eb266ea560b437c3a"
          },
          "model_module_version": "1.5.0"
        },
        "829453c7bc7146fba2d76112fba0c5c9": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "initial",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          },
          "model_module_version": "1.5.0"
        },
        "3b3ee50161c247cb9489cb529d8d6dba": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          },
          "model_module_version": "1.2.0"
        },
        "d5ded497a4f4490a8de285779dc5e501": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          },
          "model_module_version": "1.5.0"
        },
        "a4e7c79f80d4493eb266ea560b437c3a": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          },
          "model_module_version": "1.2.0"
        }
      }
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/JosephBless/Deep-Learning-Colab/blob/main/BERT_Embeddings.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0kY-xmMZ1qQy"
      },
      "source": [
        "# Second Experiment\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OuV1E-9u1qQ1"
      },
      "source": [
        "For our second experiment, we connected the Notebook to an Elasticsearch instance and compared a standard Elasticsearch operator with Embeddings we got from BERT (Bi-Directional Encoder Representation from Transformer). The Embeddings are indexed, built-in stemming methods: 'Stemmer Token Filter' and 'Hunspell Token Filter'.\n",
        "(To read details about this experiment visit our [website](https://pragmalingu.de/docs/experiments/experiment1))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5nK9TA1Z1qQ1"
      },
      "source": [
        "## Setup an Elasticsearch Instance in Google Colab"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vQYmJo3X1qQ1"
      },
      "source": [
        "Everthing to connect to Elasticsearch. For a detailed explaination, see [this  Notebook.](https://pragmalingu.de/docs/experiments/experiment1)\n",
        "Download:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "p8goeWCC1qQ2"
      },
      "source": [
        "import os\n",
        "from subprocess import Popen, PIPE, STDOUT\n",
        "# download elasticsearch\n",
        "!wget https://artifacts.elastic.co/downloads/elasticsearch/elasticsearch-7.9.1-linux-x86_64.tar.gz -q\n",
        "!tar -xzf elasticsearch-7.9.1-linux-x86_64.tar.gz\n",
        "!chown -R daemon:daemon elasticsearch-7.9.1\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8kiNBCMz1qQ2"
      },
      "source": [
        "Start a local server:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "D_rgN1tZ1qQ2",
        "outputId": "fb6cd7aa-7108-4c77-beba-2e3766d1395a"
      },
      "source": [
        "# start server\n",
        "es_server = Popen(['elasticsearch-7.9.1/bin/elasticsearch'],\n",
        "                  stdout=PIPE, stderr=STDOUT,\n",
        "                  preexec_fn=lambda: os.setuid(1)  # as daemon\n",
        "                 )\n",
        "# client-side\n",
        "!pip install elasticsearch -q\n",
        "from elasticsearch import Elasticsearch\n",
        "from datetime import datetime\n",
        "es = Elasticsearch([\"localhost:9200/\"])\n",
        "#wait a bit\n",
        "import time\n",
        "time.sleep(30)\n",
        "es.ping()  # got True"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[?25l\r\u001b[K     |█                               | 10kB 13.9MB/s eta 0:00:01\r\u001b[K     |██                              | 20kB 3.2MB/s eta 0:00:01\r\u001b[K     |███                             | 30kB 4.5MB/s eta 0:00:01\r\u001b[K     |███▉                            | 40kB 5.7MB/s eta 0:00:01\r\u001b[K     |████▉                           | 51kB 6.1MB/s eta 0:00:01\r\u001b[K     |█████▉                          | 61kB 5.6MB/s eta 0:00:01\r\u001b[K     |██████▊                         | 71kB 6.1MB/s eta 0:00:01\r\u001b[K     |███████▊                        | 81kB 4.2MB/s eta 0:00:01\r\u001b[K     |████████▊                       | 92kB 4.6MB/s eta 0:00:01\r\u001b[K     |█████████▋                      | 102kB 5.0MB/s eta 0:00:01\r\u001b[K     |██████████▋                     | 112kB 5.0MB/s eta 0:00:01\r\u001b[K     |███████████▋                    | 122kB 5.0MB/s eta 0:00:01\r\u001b[K     |████████████▌                   | 133kB 5.0MB/s eta 0:00:01\r\u001b[K     |█████████████▌                  | 143kB 5.0MB/s eta 0:00:01\r\u001b[K     |██████████████▌                 | 153kB 5.0MB/s eta 0:00:01\r\u001b[K     |███████████████▍                | 163kB 5.0MB/s eta 0:00:01\r\u001b[K     |████████████████▍               | 174kB 5.0MB/s eta 0:00:01\r\u001b[K     |█████████████████▍              | 184kB 5.0MB/s eta 0:00:01\r\u001b[K     |██████████████████▎             | 194kB 5.0MB/s eta 0:00:01\r\u001b[K     |███████████████████▎            | 204kB 5.0MB/s eta 0:00:01\r\u001b[K     |████████████████████▎           | 215kB 5.0MB/s eta 0:00:01\r\u001b[K     |█████████████████████▏          | 225kB 5.0MB/s eta 0:00:01\r\u001b[K     |██████████████████████▏         | 235kB 5.0MB/s eta 0:00:01\r\u001b[K     |███████████████████████▏        | 245kB 5.0MB/s eta 0:00:01\r\u001b[K     |████████████████████████        | 256kB 5.0MB/s eta 0:00:01\r\u001b[K     |█████████████████████████       | 266kB 5.0MB/s eta 0:00:01\r\u001b[K     |██████████████████████████      | 276kB 5.0MB/s eta 0:00:01\r\u001b[K     |███████████████████████████     | 286kB 5.0MB/s eta 0:00:01\r\u001b[K     |████████████████████████████    | 296kB 5.0MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████   | 307kB 5.0MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████  | 317kB 5.0MB/s eta 0:00:01\r\u001b[K     |██████████████████████████████▉ | 327kB 5.0MB/s eta 0:00:01\r\u001b[K     |███████████████████████████████▉| 337kB 5.0MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 348kB 5.0MB/s \n",
            "\u001b[?25h"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "True"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 2
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "gzBSBJ_81qQ3",
        "outputId": "b73da037-bf69-4cf7-f2ec-eceeccce796a"
      },
      "source": [
        "#print new index list\n",
        "create_response = es.cat.indices()\n",
        "print(create_response)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8AHEMh-f1qQ4"
      },
      "source": [
        "## Download and intialize a pretrained BERT model\n",
        "\n",
        "### Bi-Directional Encoder Representation from Transformer (BERT)\n",
        "\n",
        "BERT is an technique which uses large, pretrained neural networks - with some exceptional solutions - to get the vectors from texts. We can then use these with some similarity metrics, like cosine similarity, to compare the meaning of these texts.\n",
        "\n",
        "(By the way these networks are frequently used as a backbone, or part of an ensemble, of models to solve some NLP tasks like Question Answering, Ranking, Named Entitity Recognition, etc.)\n",
        "\n",
        "[I'm brave enough to read the paper on BERT](https://arxiv.org/abs/1810.04805)\n",
        "\n",
        "\n",
        "### How do we plan to use it?\n",
        "\n",
        " - Get embeddings (vector representations) from documents using BERT\n",
        " - Index them using the KNN algorithm included in ElasticSearch\n",
        " - Get embeddings from queries (there are relevance labels of pairs query-document) using BERT\n",
        " - Use relevance labels and ranking API from ElasticSearch to calculate metrics and compare them with classical approaches"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8Udoh89I1qQ4"
      },
      "source": [
        "### Basic Demonstration"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "2rPNYLNs1qQ5"
      },
      "source": [
        "** Download a pretrained BERT model**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 926,
          "referenced_widgets": [
            "6c05278600fe4b6da1ee0f25b22f5539",
            "38a6d5d0831f49ed87a0188cfcbfce4c",
            "a61785fe0a364969ab3d50cba4b9504d",
            "099b7f85c72f4fddaceb5d2674485857",
            "5b4c15cd098243338e685b0382abc91d",
            "7066203a1fac43f095ac16cc85d99e2d",
            "21dab7daef1f40deb940bfc841eb3210",
            "faa3b6da40364748b233e25ce23642bc"
          ]
        },
        "id": "jt_mFyoq1qQ5",
        "outputId": "b11799e4-8f61-40ec-cf33-384056225221"
      },
      "source": [
        "!pip install -U sentence-transformers\n",
        "\n",
        "from sentence_transformers import SentenceTransformer\n",
        "import torch\n",
        "from tqdm import tqdm_notebook\n",
        "\n",
        "model = SentenceTransformer('bert-base-nli-mean-tokens')\n",
        "\n",
        "# using gpu to boost inference if it's possible\n",
        "if torch.cuda.is_available():\n",
        "  model.to('cuda')\n",
        "\n",
        "print('Max Sequence Length:', model.max_seq_length)\n",
        "\n",
        "#Change the length to max possible length (based on gpu memory)\n",
        "model.max_seq_length = 364\n",
        "\n",
        "print('Max Sequence Length:', model.max_seq_length)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting sentence-transformers\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/c4/87/49dc49e13ac107ce912c2f3f3fd92252c6d4221e88d1e6c16747044a11d8/sentence-transformers-1.1.0.tar.gz (78kB)\n",
            "\u001b[K     |████████████████████████████████| 81kB 4.2MB/s \n",
            "\u001b[?25hCollecting transformers<5.0.0,>=3.1.0\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/d8/b2/57495b5309f09fa501866e225c84532d1fd89536ea62406b2181933fb418/transformers-4.5.1-py3-none-any.whl (2.1MB)\n",
            "\u001b[K     |████████████████████████████████| 2.1MB 8.9MB/s \n",
            "\u001b[?25hRequirement already satisfied, skipping upgrade: tqdm in /usr/local/lib/python3.7/dist-packages (from sentence-transformers) (4.41.1)\n",
            "Requirement already satisfied, skipping upgrade: torch>=1.6.0 in /usr/local/lib/python3.7/dist-packages (from sentence-transformers) (1.8.1+cu101)\n",
            "Requirement already satisfied, skipping upgrade: numpy in /usr/local/lib/python3.7/dist-packages (from sentence-transformers) (1.19.5)\n",
            "Requirement already satisfied, skipping upgrade: scikit-learn in /usr/local/lib/python3.7/dist-packages (from sentence-transformers) (0.22.2.post1)\n",
            "Requirement already satisfied, skipping upgrade: scipy in /usr/local/lib/python3.7/dist-packages (from sentence-transformers) (1.4.1)\n",
            "Requirement already satisfied, skipping upgrade: nltk in /usr/local/lib/python3.7/dist-packages (from sentence-transformers) (3.2.5)\n",
            "Collecting sentencepiece\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/f5/99/e0808cb947ba10f575839c43e8fafc9cc44e4a7a2c8f79c60db48220a577/sentencepiece-0.1.95-cp37-cp37m-manylinux2014_x86_64.whl (1.2MB)\n",
            "\u001b[K     |████████████████████████████████| 1.2MB 27.9MB/s \n",
            "\u001b[?25hRequirement already satisfied, skipping upgrade: importlib-metadata; python_version < \"3.8\" in /usr/local/lib/python3.7/dist-packages (from transformers<5.0.0,>=3.1.0->sentence-transformers) (3.10.1)\n",
            "Requirement already satisfied, skipping upgrade: filelock in /usr/local/lib/python3.7/dist-packages (from transformers<5.0.0,>=3.1.0->sentence-transformers) (3.0.12)\n",
            "Collecting sacremoses\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/75/ee/67241dc87f266093c533a2d4d3d69438e57d7a90abb216fa076e7d475d4a/sacremoses-0.0.45-py3-none-any.whl (895kB)\n",
            "\u001b[K     |████████████████████████████████| 901kB 36.5MB/s \n",
            "\u001b[?25hRequirement already satisfied, skipping upgrade: regex!=2019.12.17 in /usr/local/lib/python3.7/dist-packages (from transformers<5.0.0,>=3.1.0->sentence-transformers) (2019.12.20)\n",
            "Requirement already satisfied, skipping upgrade: packaging in /usr/local/lib/python3.7/dist-packages (from transformers<5.0.0,>=3.1.0->sentence-transformers) (20.9)\n",
            "Requirement already satisfied, skipping upgrade: requests in /usr/local/lib/python3.7/dist-packages (from transformers<5.0.0,>=3.1.0->sentence-transformers) (2.23.0)\n",
            "Collecting tokenizers<0.11,>=0.10.1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/ae/04/5b870f26a858552025a62f1649c20d29d2672c02ff3c3fb4c688ca46467a/tokenizers-0.10.2-cp37-cp37m-manylinux2010_x86_64.whl (3.3MB)\n",
            "\u001b[K     |████████████████████████████████| 3.3MB 40.0MB/s \n",
            "\u001b[?25hRequirement already satisfied, skipping upgrade: typing-extensions in /usr/local/lib/python3.7/dist-packages (from torch>=1.6.0->sentence-transformers) (3.7.4.3)\n",
            "Requirement already satisfied, skipping upgrade: joblib>=0.11 in /usr/local/lib/python3.7/dist-packages (from scikit-learn->sentence-transformers) (1.0.1)\n",
            "Requirement already satisfied, skipping upgrade: six in /usr/local/lib/python3.7/dist-packages (from nltk->sentence-transformers) (1.15.0)\n",
            "Requirement already satisfied, skipping upgrade: zipp>=0.5 in /usr/local/lib/python3.7/dist-packages (from importlib-metadata; python_version < \"3.8\"->transformers<5.0.0,>=3.1.0->sentence-transformers) (3.4.1)\n",
            "Requirement already satisfied, skipping upgrade: click in /usr/local/lib/python3.7/dist-packages (from sacremoses->transformers<5.0.0,>=3.1.0->sentence-transformers) (7.1.2)\n",
            "Requirement already satisfied, skipping upgrade: pyparsing>=2.0.2 in /usr/local/lib/python3.7/dist-packages (from packaging->transformers<5.0.0,>=3.1.0->sentence-transformers) (2.4.7)\n",
            "Requirement already satisfied, skipping upgrade: certifi>=2017.4.17 in /usr/local/lib/python3.7/dist-packages (from requests->transformers<5.0.0,>=3.1.0->sentence-transformers) (2020.12.5)\n",
            "Requirement already satisfied, skipping upgrade: idna<3,>=2.5 in /usr/local/lib/python3.7/dist-packages (from requests->transformers<5.0.0,>=3.1.0->sentence-transformers) (2.10)\n",
            "Requirement already satisfied, skipping upgrade: chardet<4,>=3.0.2 in /usr/local/lib/python3.7/dist-packages (from requests->transformers<5.0.0,>=3.1.0->sentence-transformers) (3.0.4)\n",
            "Requirement already satisfied, skipping upgrade: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.7/dist-packages (from requests->transformers<5.0.0,>=3.1.0->sentence-transformers) (1.24.3)\n",
            "Building wheels for collected packages: sentence-transformers\n",
            "  Building wheel for sentence-transformers (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for sentence-transformers: filename=sentence_transformers-1.1.0-cp37-none-any.whl size=119615 sha256=c2d0ae1e2aa389e8d116904dc7dd70e89dd2da6a9671fe109589ec060e827082\n",
            "  Stored in directory: /root/.cache/pip/wheels/84/cb/21/1066bff3027215c760ca14a198f698bca8fccb92e33e2327eb\n",
            "Successfully built sentence-transformers\n",
            "Installing collected packages: sacremoses, tokenizers, transformers, sentencepiece, sentence-transformers\n",
            "Successfully installed sacremoses-0.0.45 sentence-transformers-1.1.0 sentencepiece-0.1.95 tokenizers-0.10.2 transformers-4.5.1\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "6c05278600fe4b6da1ee0f25b22f5539",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=0.0, max=405234788.0), HTML(value='')))"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Max Sequence Length: 128\n",
            "Max Sequence Length: 364\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DsD36mJu1qQ5"
      },
      "source": [
        "sentences = ['This framework generates embeddings for each input sentence',\n",
        "    'Sentences are passed as a list of string.',\n",
        "    'The quick brown fox jumps over the lazy dog.']\n",
        "sentence_embeddings = model.encode(sentences)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "AGbHzJN21qQ6",
        "outputId": "9d95b728-010d-4656-d463-65f847c0f4fd"
      },
      "source": [
        "for sentence, embedding in zip(sentences, sentence_embeddings):\n",
        "    print('Sentence:', sentence)\n",
        "    print('Embedding:', list(embedding[:5]) + ['...'])\n",
        "    print('Embedding\\'s length:', len(embedding))\n",
        "    print('')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Sentence: This framework generates embeddings for each input sentence\n",
            "Embedding: [-0.104094714, 0.52747625, 1.1797732, 0.19436583, 0.15911391, '...']\n",
            "Embedding's length: 768\n",
            "\n",
            "Sentence: Sentences are passed as a list of string.\n",
            "Embedding: [-0.13118386, -0.17390253, 1.1052185, 0.2485416, 0.0410018, '...']\n",
            "Embedding's length: 768\n",
            "\n",
            "Sentence: The quick brown fox jumps over the lazy dog.\n",
            "Embedding: [-0.74899274, 0.71891713, -1.0394576, 0.06408766, 0.050704937, '...']\n",
            "Embedding's length: 768\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ATb-_3cQ1qQ6"
      },
      "source": [
        "**Indexing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WeGY-oKC1qQ6",
        "outputId": "19ff6799-0e9b-4bc4-f2c0-21728fa99400"
      },
      "source": [
        "settings = {\n",
        "  \"mappings\": {\n",
        "    \"properties\": {\n",
        "      \"bert_vector\": {\n",
        "         \"type\": \"dense_vector\",\n",
        "        \"dims\": 768\n",
        "      }\n",
        "    }\n",
        "  }\n",
        "}\n",
        "\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "toy_index = \"bert-toy_index\"\n",
        "es.indices.delete(index=toy_index, ignore=[400, 404])\n",
        "es.indices.create(toy_index, body=settings)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True, 'index': 'bert-toy_index', 'shards_acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 124,
          "referenced_widgets": [
            "ac661179add94abfa964cec7fda930e9",
            "dd5caeaee01a4dc0bb7fe2eb966405f9",
            "e108dc18b08e424cb3a7c7c48f536dff",
            "c8d3c5675def444bac6d8029abf11432",
            "829453c7bc7146fba2d76112fba0c5c9",
            "3b3ee50161c247cb9489cb529d8d6dba",
            "d5ded497a4f4490a8de285779dc5e501",
            "a4e7c79f80d4493eb266ea560b437c3a"
          ]
        },
        "id": "U70-8wvA1qQ6",
        "outputId": "19066e04-462b-4e39-b064-f66ee6262cf2"
      },
      "source": [
        "from tqdm import tqdm_notebook\n",
        "\n",
        "model.eval()\n",
        "\n",
        "for i, sentence in tqdm_notebook(enumerate(sentences)):\n",
        "  with torch.no_grad():\n",
        "    if torch.cuda.is_available():\n",
        "      torch.cuda.ipc_collect()\n",
        "      torch.cuda.empty_cache()\n",
        "    es.index(\n",
        "      index=toy_index,\n",
        "      id=i,\n",
        "      body={\n",
        "          'bert_vector': model.encode(sentence),\n",
        "          'text': sentence\n",
        "          }\n",
        "    )"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/ipykernel_launcher.py:5: TqdmDeprecationWarning: This function will be removed in tqdm==5.0.0\n",
            "Please use `tqdm.notebook.tqdm` instead of `tqdm.tqdm_notebook`\n",
            "  \"\"\"\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "ac661179add94abfa964cec7fda930e9",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UAmjUNdX1qQ7"
      },
      "source": [
        "**Searching**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "VramsjG_1qQ7",
        "outputId": "e100c661-6166-4c9e-a4f2-8f51805bb10a"
      },
      "source": [
        "test_str = \"Where is the fox?\"\n",
        "\n",
        "#test query knn-search\n",
        "res = es.search(\n",
        "    index=\"bert-toy_index\",\n",
        "    body={\n",
        "        \"query\" : {\n",
        "            \"script_score\": {\n",
        "                \"query\": {\"match_all\": {}},\n",
        "                \"script\": {\n",
        "                    \"source\": \"cosineSimilarity(params.query_vector, doc['bert_vector']) + 1.0\",\n",
        "                    \"params\": {\"query_vector\": list(model.encode(test_str).astype(float))}\n",
        "                                                   }}}})\n",
        "\n",
        "\n",
        "print(\"Got %d Hits:\\n\\n\" % res['hits']['total']['value'])\n",
        "for hit in res['hits']['hits']:\n",
        "  print(f\"Cosine similarity score: {hit['_score']}  \\nid: {hit['_id']}\\ntext: {hit['_source']['text']}\\n\\n\")\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Got 0 Hits:\n",
            "\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "N4UqDYWe1qQ8",
        "outputId": "d2d910f2-0cf7-4f1e-d010-a417ae1a8fbe"
      },
      "source": [
        "# Deleting index\n",
        "es.indices.delete(index=toy_index, ignore=[400, 404])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7IJId6Vt1qQ8"
      },
      "source": [
        "## Parse Data\n",
        "\n",
        "Get different corpora, format them, and feed them to Elasticsearch"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "o2ooC4f-1qQ9"
      },
      "source": [
        "### ADI Corpus"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "K00t3Ilc1qQ9"
      },
      "source": [
        "You can get the corpus from [this link](http://ir.dcs.gla.ac.uk/resources/test_collections/adi/).  <br>\n",
        "For detailed information about the parsing of this corpus, look at [ this Notebook](https://colab.research.google.com/github/pragmalingu/private_experiments/blob/adi_corpus/ADICorpus.ipynb) or, for parsing in generel, read [this guide](https://)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VFp4JTeJ1qQ9"
      },
      "source": [
        "**Parsing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yoVKvOYc1qQ9",
        "outputId": "4b62bf5c-5159-439d-f3c8-1df5ae38b57b"
      },
      "source": [
        "# download and unzip data\n",
        "!wget http://ir.dcs.gla.ac.uk/resources/test_collections/adi/adi.tar.gz\n",
        "!tar -xf adi.tar.gz\n",
        "\n",
        "# set paths to the dowloaded data as variables\n",
        "PATH_TO_ADI_TXT = '/content/ADI.ALL'\n",
        "PATH_TO_ADI_QRY = '/content/ADI.QRY'\n",
        "PATH_TO_ADI_REL = '/content/ADI.REL'\n",
        "\n",
        "from collections import defaultdict\n",
        "import re\n",
        "import json\n",
        "from io import StringIO\n",
        "import numpy as np\n",
        "\n",
        "\n",
        "# get the text and query files\n",
        "\n",
        "ID_marker = re.compile('\\.I')\n",
        "\n",
        "def get_data(PATH_TO_FILE, marker):\n",
        "  \"\"\"\n",
        "  Reads file and spilts text into entries at the ID marker '.I'.\n",
        "  First entry is empty, so it's removed.\n",
        "  'marker' contains the regex at which we want to split\n",
        "  \"\"\"\n",
        "  with open (PATH_TO_FILE,'r') as f:\n",
        "    text = f.read().replace('\\n',\" \")\n",
        "    lines = re.split(marker,text)\n",
        "    lines.pop(0)\n",
        "  return lines\n",
        "\n",
        "adi_txt_list = get_data(PATH_TO_ADI_TXT, ID_marker)\n",
        "adi_qry_list = get_data(PATH_TO_ADI_QRY, ID_marker)\n",
        "\n",
        "# process text file\n",
        "\n",
        "adi_title_start = re.compile('\\.T')\n",
        "adi_author_start = re.compile('\\.A')\n",
        "adi_text_start = re.compile('\\.W')\n",
        "\n",
        "adi_txt_data = defaultdict(dict)\n",
        "\n",
        "for line in adi_txt_list:\n",
        "  entries = re.split(adi_title_start,line,1)\n",
        "  id = entries[0].strip()\n",
        "  no_id = entries[1]\n",
        "  if len(re.split(adi_author_start, no_id,1)) > 1:\n",
        "    no_id_entries = re.split(adi_author_start, no_id,1)\n",
        "    adi_txt_data[id]['title'] = no_id_entries[0]\n",
        "    no_title = no_id_entries[1]\n",
        "    no_title_entries = re.split(adi_text_start, no_title)\n",
        "    adi_txt_data[id]['author'] = no_title_entries[0]\n",
        "    adi_txt_data[id]['text'] = no_title_entries[1]\n",
        "  else:\n",
        "    no_id_entries = re.split(adi_text_start, no_id)\n",
        "    adi_txt_data[id]['title'] = no_id_entries[0]\n",
        "    adi_txt_data[id]['text'] = no_id_entries[1]\n",
        "\n",
        "\n",
        "# process query file\n",
        "\n",
        "qry_chunk_start = re.compile('\\.W')\n",
        "adi_qry_rel_data = defaultdict(dict)\n",
        "\n",
        "for n in range(0,len(adi_qry_list)):\n",
        "  line = adi_qry_list[n]\n",
        "  _ , question = re.split(qry_chunk_start,line)\n",
        "  adi_qry_rel_data[n+1]['question'] = question\n",
        "\n",
        "# process relevance assesment\n",
        "\n",
        "adi_rel_data = open(PATH_TO_ADI_REL)\n",
        "adi_np = np.loadtxt(adi_rel_data, dtype=float)\n",
        "\n",
        "adi_rel = defaultdict(list)\n",
        "for row in adi_np:\n",
        "  adi_rel[int(row[0])].append(int(row[1]))\n",
        "\n",
        "for id, rels in adi_rel.items():\n",
        "  adi_qry_rel_data[id]['relevance_assessments'] = rels\n",
        "\n",
        "# transform text to sentences for BERT text search\n",
        "\n",
        "import nltk\n",
        "from nltk import tokenize\n",
        "from nltk.tokenize import sent_tokenize\n",
        "nltk.download('punkt')\n",
        "\n",
        "def text_to_sentences(string):\n",
        "  sentences = tokenize.sent_tokenize(string)\n",
        "  return sentences\n",
        "\n",
        "for ID, doc in adi_txt_data.items():\n",
        "  text = adi_txt_data[ID]['text']\n",
        "  adi_txt_data[ID]['sentences'] = text_to_sentences(text)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-04-29 11:21:06--  http://ir.dcs.gla.ac.uk/resources/test_collections/adi/adi.tar.gz\n",
            "Resolving ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)... 130.209.240.253\n",
            "Connecting to ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)|130.209.240.253|:80... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 17307 (17K) [application/gzip]\n",
            "Saving to: ‘adi.tar.gz’\n",
            "\n",
            "adi.tar.gz          100%[===================>]  16.90K  --.-KB/s    in 0.1s    \n",
            "\n",
            "2021-04-29 11:21:06 (161 KB/s) - ‘adi.tar.gz’ saved [17307/17307]\n",
            "\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Unzipping tokenizers/punkt.zip.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qfrjdQYE1qQ-"
      },
      "source": [
        "**Indexing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "PDozQgtC1qQ-",
        "outputId": "2d324ae2-dfbf-420c-d948-bc2871e7ce43"
      },
      "source": [
        "# Elasticsearch settings, without using Opendistro\n",
        "# Every index has title, text and vectors to fill\n",
        "settings_nest = {\n",
        "  \"mappings\": {\n",
        "    \"properties\": {\n",
        "      \"text_vector\": {\n",
        "        \"type\": \"nested\",\n",
        "        \"properties\": {\n",
        "          \"vector\": {\n",
        "            \"type\": \"dense_vector\",\n",
        "            \"dims\": 768\n",
        "          }\n",
        "        }\n",
        "      },\n",
        "     \"text\": {\n",
        "            \"type\" : \"text\"\n",
        "          },\n",
        "          \"title_vector\": {\n",
        "           \"type\": \"dense_vector\",\n",
        "           \"dims\": 768\n",
        "            },\n",
        "      \"title\": {\n",
        "              \"type\": \"text\"\n",
        "              },\n",
        "    }\n",
        "  }\n",
        "}\n",
        "\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "adi_index = \"adi-corpus\"\n",
        "es.indices.delete(index=adi_index, ignore=[400, 404])\n",
        "es.indices.create(adi_index)\n",
        "\n",
        "adi_index_sent = \"bert-sentence-adi-corpus\"\n",
        "es.indices.delete(index=adi_index_sent, ignore=[400, 404])\n",
        "es.indices.create(adi_index_sent, body=settings_nest)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True,\n",
              " 'index': 'bert-sentence-adi-corpus',\n",
              " 'shards_acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "1USBxsME1qQ_",
        "outputId": "ac44263f-9f3e-40a2-cb9f-5ef3ddc5890d"
      },
      "source": [
        "for ID, doc_data in adi_txt_data.items():\n",
        "    # index for match query\n",
        "    es.index(index=adi_index, id=ID, body=doc_data)\n",
        "    # index for BERT search\n",
        "    es.index(\n",
        "      index=adi_index_sent,\n",
        "      id=ID,\n",
        "      body={\n",
        "          'text_vector': [{\"vector\": model.encode(text)} for text in adi_txt_data[ID]['sentences']],\n",
        "          'text': adi_txt_data[ID]['text'],\n",
        "          'title_vector': model.encode(adi_txt_data[ID]['title']),\n",
        "          'title': adi_txt_data[ID]['title'],\n",
        "          }\n",
        "    )\n",
        "\n",
        "#print new index list\n",
        "create_response = es.cat.indices()\n",
        "print(create_response)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "yellow open adi-corpus               VhHKO45mTSuy4SRdtN199Q 1 1  78 0 221.3kb 221.3kb\n",
            "yellow open bert-sentence-adi-corpus zX-Lk5SnQXOVzBcpw5IusQ 1 1 254 0   4.2mb   4.2mb\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3LDqN-0M1qQ_"
      },
      "source": [
        "### CACM Corpus"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Qchg1nve1qQ_"
      },
      "source": [
        "You can get the corpus from [this link](http://ir.dcs.gla.ac.uk/resources/test_collections/cacm/).  <br>\n",
        "For detailed information about the format of the files, see the PragmaLingu [ Benchmarks](https://pragmalingu.de/docs/benchmarks/overview)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ONL1thG_1qQ_"
      },
      "source": [
        "**Parsing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "mwG3gKXZ1qQ_",
        "outputId": "be2a253d-8163-42c6-b394-53900d14daaf"
      },
      "source": [
        "# download and unzip data\n",
        "!wget http://ir.dcs.gla.ac.uk/resources/test_collections/cacm/cacm.tar.gz\n",
        "!tar -xf cacm.tar.gz\n",
        "\n",
        "# set paths to the dowloaded data as variablesDownload and unzip data.\n",
        "\n",
        "PATH_TO_CACM_TXT = '/content/cacm.all'\n",
        "PATH_TO_CACM_QRY = '/content/query.text'\n",
        "PATH_TO_CACM_REL = '/content/qrels.text'\n",
        "PATH_TO_CACM_COMMON_WORDS = '/content/common_words'\n",
        "\n",
        "from collections import defaultdict\n",
        "import re\n",
        "import json\n",
        "from io import StringIO\n",
        "import numpy as np\n",
        "\n",
        "# get the text and query files\n",
        "\n",
        "ID_marker = re.compile('^\\.I',re.MULTILINE)\n",
        "\n",
        "def get_data(PATH_TO_FILE, marker):\n",
        "  \"\"\"\n",
        "  Reads file and spilts text into entries at the ID marker '.I'.\n",
        "  First entry is empty, so it's removed.\n",
        "  'marker' contains the regex at which we want to split\n",
        "  \"\"\"\n",
        "  with open (PATH_TO_FILE,'r') as f:\n",
        "    text = f.read()\n",
        "    lines = re.split(marker,text)\n",
        "    lines.pop(0)\n",
        "  return lines\n",
        "\n",
        "cacm_txt_list = get_data(PATH_TO_CACM_TXT, ID_marker)\n",
        "cacm_qry_list = get_data(PATH_TO_CACM_QRY, ID_marker)\n",
        "\n",
        "# process text file\n",
        "cacm_chunk_title = re.compile('\\.[T]\\n')\n",
        "cacm_chunk_markers = re.compile('^\\.', re.MULTILINE)\n",
        "cacm_chunk_txt = re.compile('W\\n')\n",
        "cacm_chunk_pub = re.compile('B\\n')\n",
        "cacm_chunk_author =re.compile('A\\n')\n",
        "cacm_chunk_add = re.compile('N\\n')\n",
        "cacm_chunk_cross = re.compile('X\\n')\n",
        "\n",
        "cacm_txt_data = defaultdict(dict)\n",
        "\n",
        "for line in cacm_txt_list:\n",
        "    entries = re.split(cacm_chunk_title, line)\n",
        "    id = entries[0].strip()  # save id\n",
        "    no_id = entries[1]\n",
        "    list_of_entries = re.split(cacm_chunk_markers, no_id)\n",
        "    cacm_txt_data[id]['title'] = list_of_entries[0].strip()\n",
        "    for entry in list_of_entries:\n",
        "        if len(re.split(cacm_chunk_txt, entry)) == 2: # check for text\n",
        "            cacm_txt_data[id]['text'] = re.split(cacm_chunk_txt, entry)[1]\n",
        "        if len(re.split(cacm_chunk_pub, entry)) == 2: # check for publication date\n",
        "          cacm_txt_data[id]['publication_date'] = re.split(cacm_chunk_pub, entry)[1]\n",
        "        if len(re.split(cacm_chunk_author, entry)) == 2: # check for author\n",
        "            cacm_txt_data[id]['author'] = re.split(cacm_chunk_author, entry)[1]\n",
        "        if len(re.split(cacm_chunk_add, entry)) == 2: # check for additional date\n",
        "            cacm_txt_data[id]['add_date'] = re.split(cacm_chunk_add, entry)[1]\n",
        "        if len(re.split(cacm_chunk_cross, entry)) == 2:  # check for cross-references\n",
        "            cacm_txt_data[id]['cross-references'] = re.split(cacm_chunk_cross, entry)[1]\n",
        "\n",
        "# process the query data\n",
        "qry_chunk_start = re.compile('^\\.W\\n',re.MULTILINE)\n",
        "qry_chunk_end = re.compile('^\\.N\\n',re.MULTILINE)\n",
        "\n",
        "cacm_qry_rel_data = defaultdict(dict)\n",
        "\n",
        "for n in range(0,len(cacm_qry_list)):\n",
        "  line = cacm_qry_list[n]\n",
        "  _ , question = re.split(qry_chunk_start,line)\n",
        "  query, _ = re.split(qry_chunk_end, question)\n",
        "  cacm_qry_rel_data[n+1]['question'] = query\n",
        "\n",
        "# process the relevance assesment data\n",
        "cacm_rel = defaultdict(list)\n",
        "\n",
        "with open (PATH_TO_CACM_REL,'r') as f:\n",
        "  for line in f:\n",
        "    line = re.split(' ',line)\n",
        "    cacm_rel[int(line[0])].append(line[1])\n",
        "\n",
        "for id, rels in cacm_rel.items():\n",
        "  cacm_qry_rel_data[id]['relevance_assessments'] = [int(r) for r in rels]\n",
        "\n",
        "\n",
        "# transform text to sentences for BERT text search\n",
        "\n",
        "import nltk\n",
        "from nltk import tokenize\n",
        "from nltk.tokenize import sent_tokenize\n",
        "nltk.download('punkt')\n",
        "\n",
        "def text_to_sentences(string):\n",
        "  sentences = tokenize.sent_tokenize(string)\n",
        "  return sentences\n",
        "\n",
        "for ID, doc in cacm_txt_data.items():\n",
        "  if 'text' in cacm_txt_data[str(ID)]:\n",
        "    text = cacm_txt_data[ID]['text']\n",
        "    cacm_txt_data[ID]['sentences'] = text_to_sentences(text)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-04-29 11:21:19--  http://ir.dcs.gla.ac.uk/resources/test_collections/cacm/cacm.tar.gz\n",
            "Resolving ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)... 130.209.240.253\n",
            "Connecting to ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)|130.209.240.253|:80... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 705665 (689K) [application/gzip]\n",
            "Saving to: ‘cacm.tar.gz’\n",
            "\n",
            "cacm.tar.gz         100%[===================>] 689.13K  1.26MB/s    in 0.5s    \n",
            "\n",
            "2021-04-29 11:21:19 (1.26 MB/s) - ‘cacm.tar.gz’ saved [705665/705665]\n",
            "\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "F92TVp2Y1qRA"
      },
      "source": [
        "**Indexing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "tB1k5TaK1qRB",
        "outputId": "f939557c-52d1-42c1-9443-63e8207250d0"
      },
      "source": [
        "# Elasticsearch settings, without using Opendistro\n",
        "# Every index has title, text and vectors to fill\n",
        "settings_nest = {\n",
        "  \"mappings\": {\n",
        "    \"properties\": {\n",
        "      \"text_vector\": {\n",
        "        \"type\": \"nested\",\n",
        "        \"properties\": {\n",
        "          \"vector\": {\n",
        "            \"type\": \"dense_vector\",\n",
        "            \"dims\": 768\n",
        "          }\n",
        "        }\n",
        "      },\n",
        "     \"text\": {\n",
        "            \"type\" : \"text\"\n",
        "          },\n",
        "          \"title_vector\": {\n",
        "           \"type\": \"dense_vector\",\n",
        "           \"dims\": 768\n",
        "            },\n",
        "      \"title\": {\n",
        "              \"type\": \"text\"\n",
        "              },\n",
        "    }\n",
        "  }\n",
        "}\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "cacm_index= \"cacm-corpus\"\n",
        "es.indices.delete(index=cacm_index, ignore=[400, 404])\n",
        "es.indices.create(cacm_index)\n",
        "\n",
        "cacm_index_sent = \"bert-sentence-cacm-corpus\"\n",
        "es.indices.delete(index=cacm_index_sent, ignore=[400, 404])\n",
        "es.indices.create(cacm_index_sent, body=settings_nest)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True,\n",
              " 'index': 'bert-sentence-cacm-corpus',\n",
              " 'shards_acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 15
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "JExpHZhV1qRB",
        "outputId": "bbd35197-ea59-427f-e573-17dade3b6164"
      },
      "source": [
        "for ID, doc_data in cacm_txt_data.items():\n",
        "    # index for match query\n",
        "    es.index(index=cacm_index, id=ID, body=doc_data)\n",
        "    # index for BERT search, not every entry has a text\n",
        "    if 'text' not in cacm_txt_data[str(ID)]:\n",
        "      es.index(\n",
        "        index=cacm_index_sent,\n",
        "        id=ID,\n",
        "        body={\n",
        "            'text_vector': [{\"vector\": model.encode(char)} for char in 'No'],\n",
        "            'text': '',\n",
        "            'title_vector': model.encode(cacm_txt_data[ID]['title']),\n",
        "            'title': cacm_txt_data[ID]['title'],\n",
        "            }\n",
        "             )\n",
        "    else:\n",
        "      es.index(\n",
        "      index=cacm_index_sent,\n",
        "      id=ID,\n",
        "      body={\n",
        "          'text_vector': [{\"vector\": model.encode(text)} for text in cacm_txt_data[ID]['sentences']],\n",
        "          'text': cacm_txt_data[ID]['text'],\n",
        "          'title_vector': model.encode(cacm_txt_data[ID]['title']),\n",
        "          'title': cacm_txt_data[ID]['title'],\n",
        "          }\n",
        "    )\n",
        "\n",
        "\n",
        "#print new index list\n",
        "create_response = es.cat.indices()\n",
        "print(create_response)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "yellow open bert-sentence-cacm-corpus wQ-lRS5ISYm1FT4vzZHnNQ 1 1 872 0 160.5mb 160.5mb\n",
            "yellow open cacm-corpus               hW80LSBXRtalOg57jYJoJw 1 1 278 0   1.6mb   1.6mb\n",
            "yellow open adi-corpus                VhHKO45mTSuy4SRdtN199Q 1 1  82 0 160.5kb 160.5kb\n",
            "yellow open bert-sentence-adi-corpus  zX-Lk5SnQXOVzBcpw5IusQ 1 1 285 0   4.3mb   4.3mb\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sL4EzDx31qRC"
      },
      "source": [
        "### CISI Corpus"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EXmlcTVW1qRC"
      },
      "source": [
        "You can get the corpus from [this link](http://ir.dcs.gla.ac.uk/resources/test_collections/cisi/).  <br>\n",
        "For detailed information about the parsing of this corpus look, at [ this Notebook](https://colab.research.google.com/github/pragmalingu/private_experiments/blob/cisi_corpus/CISICorpus.ipynb) or, for parsing in generel, read [this guide](https://)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "tavy17h-1qRC"
      },
      "source": [
        "**Parsing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nO7rvA1O1qRC",
        "outputId": "13b5287a-6a4d-4589-c808-f9bcb033aba9"
      },
      "source": [
        "# download and unzip data\n",
        "!wget http://ir.dcs.gla.ac.uk/resources/test_collections/cisi/cisi.tar.gz\n",
        "!tar -xf cisi.tar.gz\n",
        "\n",
        "# set paths to the dowloaded data as variablesDownload and unzip data.\n",
        "PATH_TO_CISI_TXT = '/content/CISI.ALL'\n",
        "PATH_TO_CISI_QRY = '/content/CISI.QRY'\n",
        "PATH_TO_CISI_REL = '/content/CISI.REL'\n",
        "\n",
        "from collections import defaultdict\n",
        "import re\n",
        "import json\n",
        "from io import StringIO\n",
        "import numpy as np\n",
        "\n",
        "# get the text and query files\n",
        "\n",
        "ID_marker = re.compile('^\\.I',re.MULTILINE)\n",
        "\n",
        "def get_data(PATH_TO_FILE, marker):\n",
        "  \"\"\"\n",
        "  Reads file and spilts text into entries at the ID marker '.I'.\n",
        "  First entry is empty, so it's removed.\n",
        "  'marker' contains the regex at which we want to split\n",
        "  \"\"\"\n",
        "  with open (PATH_TO_FILE,'r') as f:\n",
        "    text = f.read()\n",
        "    lines = re.split(marker,text)\n",
        "    lines.pop(0)\n",
        "  return lines\n",
        "\n",
        "cisi_txt_list = get_data(PATH_TO_CISI_TXT, ID_marker)\n",
        "cisi_qry_list = get_data(PATH_TO_CISI_QRY, ID_marker)\n",
        "\n",
        "# process text file\n",
        "\n",
        "cisi_title_start = re.compile('[\\n]\\.T')\n",
        "cisi_author_start = re.compile('[\\n]\\.A')\n",
        "cisi_date_start = re.compile('[\\n]\\.B')\n",
        "cisi_text_start = re.compile('[\\n]\\.W')\n",
        "cisi_cross_start = re.compile('[\\n]\\.X')\n",
        "\n",
        "cisi_txt_data = defaultdict(dict)\n",
        "\n",
        "for line in cisi_txt_list:\n",
        "  entries = re.split(cisi_title_start,line,1)\n",
        "  id = entries[0].strip()#save the id\n",
        "  no_id = entries[1]\n",
        "\n",
        "  if len(re.split(cisi_author_start, no_id)) >= 2: # is there just one author?\n",
        "    no_id_entries = re.split(cisi_author_start, no_id,1)\n",
        "    cisi_txt_data[id]['title'] = no_id_entries[0].strip() # save title\n",
        "    no_title = no_id_entries[1]\n",
        "\n",
        "    if len(re.split(cisi_date_start, no_title)) > 1: # is there a publication date?\n",
        "      no_title_entries = re.split(cisi_date_start, no_title)\n",
        "      cisi_txt_data[id]['author'] = no_title_entries[0].strip() # save athour\n",
        "      no_author = no_title_entries[1]\n",
        "      no_author_entries = re.split(cisi_text_start, no_author)\n",
        "      cisi_txt_data[id]['publication_date'] = no_author_entries[0].strip() # save publication date\n",
        "      no_author_date = no_author_entries[1]\n",
        "    else:\n",
        "      no_title_entries = re.split(cisi_text_start, no_title)\n",
        "      cisi_txt_data[id]['author'] = no_title_entries[0].strip() # save athour\n",
        "      no_author_date = no_title_entries[1]\n",
        "\n",
        "  else:\n",
        "    no_id_entries = re.split(cisi_author_start, no_id)\n",
        "    cisi_txt_data[id]['title'] = no_id_entries[0].strip() # save title\n",
        "    cisi_txt_data[id]['author'] = no_id_entries[1].strip() # save first author\n",
        "    no_title_entries = re.split(cisi_text_start, no_title)\n",
        "    cisi_txt_data[id]['author'] += ','+no_title_entries[0].strip() # save second athour\n",
        "    no_author_date = no_title_entries[1]\n",
        "\n",
        "  last_entries = re.split(cisi_cross_start, no_author_date)\n",
        "  cisi_txt_data[id]['text'] = last_entries[0].strip() # save text\n",
        "  cisi_txt_data[id]['cross-refrences'] = last_entries[1].strip() # save cross refrences\n",
        "\n",
        "# process query file\n",
        "\n",
        "qry_chunk_start = re.compile('[\\n]\\.W')\n",
        "\n",
        "cisi_qry_rel_data = defaultdict(dict)\n",
        "\n",
        "for line in cisi_qry_list:\n",
        "  id , question = re.split(qry_chunk_start,line)\n",
        "  id = re.findall('\\d{1,3}',id)[0]\n",
        "  cisi_qry_rel_data[id]['question'] = question\n",
        "\n",
        "# process the relevance assesment data\n",
        "cisi_rel = defaultdict(list)\n",
        "\n",
        "with open (PATH_TO_CISI_REL,'r') as f:\n",
        "  for line in f:\n",
        "    line = line.replace(' ','\\t')\n",
        "    line = re.split('\\t',line)\n",
        "    line = list(filter(None, line))\n",
        "    cisi_rel[str(line[0])].append(str(line[1]))\n",
        "\n",
        "for id, rels in cisi_rel.items():\n",
        "  cisi_qry_rel_data[id]['relevance_assessments'] = [int(r) for r in rels]\n",
        "\n",
        "# transform text to sentences for BERT text search\n",
        "\n",
        "import nltk\n",
        "from nltk import tokenize\n",
        "from nltk.tokenize import sent_tokenize\n",
        "nltk.download('punkt')\n",
        "\n",
        "def text_to_sentences(string):\n",
        "  sentences = tokenize.sent_tokenize(string)\n",
        "  return sentences\n",
        "\n",
        "for ID, doc in cisi_txt_data.items():\n",
        "  text = cisi_txt_data[ID]['text']\n",
        "  cisi_txt_data[ID]['sentences'] = text_to_sentences(text)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-04-29 11:27:06--  http://ir.dcs.gla.ac.uk/resources/test_collections/cisi/cisi.tar.gz\n",
            "Resolving ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)... 130.209.240.253\n",
            "Connecting to ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)|130.209.240.253|:80... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 775144 (757K) [application/gzip]\n",
            "Saving to: ‘cisi.tar.gz’\n",
            "\n",
            "cisi.tar.gz         100%[===================>] 756.98K  1.37MB/s    in 0.5s    \n",
            "\n",
            "2021-04-29 11:27:07 (1.37 MB/s) - ‘cisi.tar.gz’ saved [775144/775144]\n",
            "\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "E4cGHxrw1qRE"
      },
      "source": [
        "**Indexing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "30NGOtai1qRE",
        "outputId": "1e6a00da-3bad-4c3e-c5f2-4e8990749d3b"
      },
      "source": [
        "# Elasticsearch settings, without using Opendistro\n",
        "# Every index has title, text and vectors to fill\n",
        "settings_nest = {\n",
        "  \"mappings\": {\n",
        "    \"properties\": {\n",
        "      \"text_vector\": {\n",
        "        \"type\": \"nested\",\n",
        "        \"properties\": {\n",
        "          \"vector\": {\n",
        "            \"type\": \"dense_vector\",\n",
        "            \"dims\": 768\n",
        "          }\n",
        "        }\n",
        "      },\n",
        "     \"text\": {\n",
        "            \"type\" : \"text\"\n",
        "          },\n",
        "          \"title_vector\": {\n",
        "           \"type\": \"dense_vector\",\n",
        "           \"dims\": 768\n",
        "            },\n",
        "      \"title\": {\n",
        "              \"type\": \"text\"\n",
        "              },\n",
        "    }\n",
        "  }\n",
        "}\n",
        "\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "cisi_index = \"cisi-corpus\"\n",
        "es.indices.delete(index=cisi_index, ignore=[400, 404])\n",
        "es.indices.create(cisi_index)\n",
        "\n",
        "cisi_index_sent = \"bert-sentence-cisi-corpus\"\n",
        "es.indices.delete(index=cisi_index_sent, ignore=[400, 404])\n",
        "es.indices.create(cisi_index_sent, body=settings_nest)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True,\n",
              " 'index': 'bert-sentence-cisi-corpus',\n",
              " 'shards_acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WlbaCvmg1qRE",
        "outputId": "80747c51-d880-4b32-adfe-5842ee2f1af7"
      },
      "source": [
        "for ID, doc_data in cisi_txt_data.items():\n",
        "    # Index for match query\n",
        "    es.index(index=cisi_index, id=ID, body=doc_data)\n",
        "      # Index for BERT search\n",
        "    es.index(\n",
        "      index=cisi_index_sent,\n",
        "      id=ID,\n",
        "      body={\n",
        "          'text_vector': [{\"vector\": model.encode(text)} for text in cisi_txt_data[ID]['sentences']],\n",
        "          'text': cisi_txt_data[ID]['text'],\n",
        "          'title_vector': model.encode(cisi_txt_data[ID]['title']),\n",
        "          'title': cisi_txt_data[ID]['title'],\n",
        "          }\n",
        "    )\n",
        "\n",
        "#print new index list\n",
        "create_response = es.cat.indices()\n",
        "print(create_response)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "yellow open bert-sentence-cacm-corpus wQ-lRS5ISYm1FT4vzZHnNQ 1 1  872 0 160.5mb 160.5mb\n",
            "yellow open cacm-corpus               hW80LSBXRtalOg57jYJoJw 1 1  278 0   1.6mb   1.6mb\n",
            "yellow open cisi-corpus               bQuJ-NwwTmC1xWgiEQFmrw 1 1  180 0   2.4mb   2.4mb\n",
            "yellow open adi-corpus                VhHKO45mTSuy4SRdtN199Q 1 1   82 0 160.5kb 160.5kb\n",
            "yellow open bert-sentence-adi-corpus  zX-Lk5SnQXOVzBcpw5IusQ 1 1  285 0   4.3mb   4.3mb\n",
            "yellow open bert-sentence-cisi-corpus xCRbxtYUSEGFZ8f1e4nmOg 1 1 1063 0  95.4mb  95.4mb\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XnXjwUWi1qRE"
      },
      "source": [
        "### Cranfield Corpus\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BGmp2cwa1qRF"
      },
      "source": [
        "You can get the corpus from [this link](http://ir.dcs.gla.ac.uk/resources/test_collections/cran/).  <br>\n",
        "For detailed information about the parsing of this corpus, look at [ this Notebook](https://colab.research.google.com/github/pragmalingu/private_experiments/blob/cranfield_corpus/CranfieldCorpus.ipynb) or, for parsing in generel, read [this guide](https://)."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mtG__q4r1qRF"
      },
      "source": [
        "**Parsing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wdFj4OTt1qRF",
        "outputId": "63974534-355a-499f-f222-412e62a4d10a"
      },
      "source": [
        "# download and unzip data\n",
        "!wget http://ir.dcs.gla.ac.uk/resources/test_collections/cran/cran.tar.gz\n",
        "!tar -xf cran.tar.gz\n",
        "\n",
        "# set paths to the dowloaded data as variables\n",
        "PATH_TO_CRAN_TXT = '/content/cran.all.1400'\n",
        "PATH_TO_CRAN_QRY = '/content/cran.qry'\n",
        "PATH_TO_CRAN_REL = '/content/cranqrel'\n",
        "\n",
        "from collections import defaultdict\n",
        "import re\n",
        "import json\n",
        "from io import StringIO\n",
        "import numpy as np\n",
        "\n",
        "\n",
        "# get the text entries from the text and query files\n",
        "\n",
        "ID_marker = re.compile('\\.I')\n",
        "\n",
        "def get_data(PATH_TO_FILE, marker):\n",
        "  \"\"\"\n",
        "  Reads file and spilts text into entries at the ID marker '.I'.\n",
        "  First entry is empty, so it's removed.\n",
        "  'marker' contains the regex at which we want to split\n",
        "  \"\"\"\n",
        "  with open (PATH_TO_FILE,'r') as f:\n",
        "    text = f.read().replace('\\n',\" \")\n",
        "    lines = re.split(marker,text)\n",
        "    lines.pop(0)\n",
        "  return lines\n",
        "\n",
        "cran_txt_list = get_data(PATH_TO_CRAN_TXT, ID_marker)\n",
        "cran_qry_list = get_data(PATH_TO_CRAN_QRY, ID_marker)\n",
        "\n",
        "# process text file\n",
        "\n",
        "cran_chunk_start = re.compile('\\.[A,B,T,W]')\n",
        "cran_txt_data = defaultdict(dict)\n",
        "\n",
        "for line in cran_txt_list:\n",
        "  entries= re.split(cran_chunk_start,line)\n",
        "  id = entries[0].strip()\n",
        "  title = entries[1]\n",
        "  author = entries[2]\n",
        "  publication_date = entries[3]\n",
        "  text = entries[4:]\n",
        "  cran_txt_data[id]['title'] = title\n",
        "  cran_txt_data[id]['author'] = author\n",
        "  cran_txt_data[id]['publication_date'] = publication_date\n",
        "  cran_txt_data[id]['text'] = ''.join(text)\n",
        "\n",
        "# process query file\n",
        "qry_chunk_start = re.compile('\\.W')\n",
        "cran_qry_rel_data = defaultdict(dict)\n",
        "\n",
        "for n in range(0,len(cran_qry_list)):\n",
        "  line = cran_qry_list[n]\n",
        "  _ , question = re.split(qry_chunk_start,line)\n",
        "  cran_qry_rel_data[n+1]['question'] = question\n",
        "\n",
        "# process relevance assesments without rating\n",
        "cran_rel = defaultdict(list)\n",
        "\n",
        "with open (PATH_TO_CRAN_REL,'r') as f:\n",
        "  for line in f:\n",
        "    line = re.split(' ',line)\n",
        "    cran_rel[int(line[0])].append(line[1])\n",
        "\n",
        "for id, rels in cran_rel.items():\n",
        "  cran_qry_rel_data[id]['relevance_assessments'] = [int(r) for r in rels]\n",
        "\n",
        "# transform text to sentences for BERT text search\n",
        "import nltk\n",
        "from nltk import tokenize\n",
        "from nltk.tokenize import sent_tokenize\n",
        "nltk.download('punkt')\n",
        "\n",
        "def text_to_sentences(string):\n",
        "  sentences = tokenize.sent_tokenize(string)\n",
        "  return sentences\n",
        "\n",
        "for ID, doc in cran_txt_data.items():\n",
        "  if 'text' in cran_txt_data[str(ID)]:\n",
        "    text = cran_txt_data[ID]['text']\n",
        "    cran_txt_data[ID]['sentences'] = text_to_sentences(text)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-04-29 11:30:24--  http://ir.dcs.gla.ac.uk/resources/test_collections/cran/cran.tar.gz\n",
            "Resolving ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)... 130.209.240.253\n",
            "Connecting to ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)|130.209.240.253|:80... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 506960 (495K) [application/gzip]\n",
            "Saving to: ‘cran.tar.gz’\n",
            "\n",
            "cran.tar.gz         100%[===================>] 495.08K   940KB/s    in 0.5s    \n",
            "\n",
            "2021-04-29 11:30:25 (940 KB/s) - ‘cran.tar.gz’ saved [506960/506960]\n",
            "\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DCHvjm9T1qRG"
      },
      "source": [
        "**Indexing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6j6GHj9Y1qRG",
        "outputId": "29061b0d-461e-4dc1-aee0-ac1129cb6572"
      },
      "source": [
        "# Elasticsearch settings, without using Opendistro\n",
        "# Every index has title, text and vectors to fill\n",
        "settings_nest = {\n",
        "  \"mappings\": {\n",
        "    \"properties\": {\n",
        "      \"text_vector\": {\n",
        "        \"type\": \"nested\",\n",
        "        \"properties\": {\n",
        "          \"vector\": {\n",
        "            \"type\": \"dense_vector\",\n",
        "            \"dims\": 768\n",
        "          }\n",
        "        }\n",
        "      },\n",
        "     \"text\": {\n",
        "            \"type\" : \"text\"\n",
        "          },\n",
        "    \"title_vector\": {\n",
        "           \"type\": \"dense_vector\",\n",
        "           \"dims\": 768\n",
        "            },\n",
        "      \"title\": {\n",
        "              \"type\": \"text\"\n",
        "              },\n",
        "    }\n",
        "  }\n",
        "}\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "\n",
        "\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "\n",
        "cran_index = \"cranfield-corpus\"\n",
        "es.indices.delete(index=cran_index, ignore=[400, 404])\n",
        "es.indices.create(cran_index)\n",
        "\n",
        "cran_index_sent = \"bert-sentence-cranfield-corpus\"\n",
        "es.indices.delete(index=cran_index_sent, ignore=[400, 404])\n",
        "es.indices.create(cran_index_sent, body=settings_nest)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True,\n",
              " 'index': 'bert-sentence-cranfield-corpus',\n",
              " 'shards_acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 21
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "BRtsXtzV1qRG",
        "outputId": "89651f89-d890-4234-f643-ccaee48df4d2"
      },
      "source": [
        "#index documents, see https://elasticsearch-py.readthedocs.io/en/master/#example-usage\n",
        "for ID, doc_data in cran_txt_data.items():\n",
        "      # Index for match query\n",
        "      es.index(index=cran_index, id=ID, body=doc_data)\n",
        "      # Index for BERT search\n",
        "      es.index(\n",
        "      index=cran_index_sent,\n",
        "      id=ID,\n",
        "      body={\n",
        "          'text_vector': [{\"vector\": model.encode(text)} for text in cran_txt_data[ID]['sentences']],\n",
        "          'text': cran_txt_data[ID]['text'],\n",
        "          'title_vector': model.encode(cran_txt_data[ID]['title']),\n",
        "          'title': cran_txt_data[ID]['title']\n",
        "          }\n",
        "    )\n",
        "\n",
        "\n",
        "#print new index list\n",
        "create_response = es.cat.indices()\n",
        "print(create_response)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "yellow open bert-sentence-cacm-corpus      wQ-lRS5ISYm1FT4vzZHnNQ 1 1 13353 0 191.4mb 191.4mb\n",
            "yellow open cranfield-corpus               j5RF5ISFSfyqm3hE-_TCBw 1 1   143 0   1.9mb   1.9mb\n",
            "yellow open bert-sentence-cranfield-corpus _VSu6MP3ToagtcTR3APXXw 1 1  1163 0   141mb   141mb\n",
            "yellow open cacm-corpus                    hW80LSBXRtalOg57jYJoJw 1 1  3204 0   4.9mb   4.9mb\n",
            "yellow open cisi-corpus                    bQuJ-NwwTmC1xWgiEQFmrw 1 1   180 0   2.4mb   2.4mb\n",
            "yellow open adi-corpus                     VhHKO45mTSuy4SRdtN199Q 1 1    82 0 160.5kb 160.5kb\n",
            "yellow open bert-sentence-cisi-corpus      xCRbxtYUSEGFZ8f1e4nmOg 1 1  1063 0  96.4mb  96.4mb\n",
            "yellow open bert-sentence-adi-corpus       zX-Lk5SnQXOVzBcpw5IusQ 1 1   285 0   4.3mb   4.3mb\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nyxRD3971qRH"
      },
      "source": [
        "### LISA Corpus"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "oOFiWiXZ1qRH"
      },
      "source": [
        "You can get the corpus from [this link](http://ir.dcs.gla.ac.uk/resources/test_collections/lisa/).  <br>\n",
        "For detailed information about the format of the files, see the PragmaLingu [ Benchmarks](https://pragmalingu.de/docs/benchmarks/overview)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sxUF6ocG1qRH"
      },
      "source": [
        "**Parsing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "JRDtQI0o1qRI",
        "outputId": "9fdffca2-37fd-4fb8-a6cf-fb652975a2d1"
      },
      "source": [
        "# download and unzip data\n",
        "!wget http://ir.dcs.gla.ac.uk/resources/test_collections/lisa/lisa.tar.gz\n",
        "!tar -xf lisa.tar.gz\n",
        "\n",
        "\n",
        "# set paths to the dowloaded data as variablesDownload and unzip data.\n",
        "\n",
        "PATH_TO_LISA_TXT = '/content/'\n",
        "PATH_TO_LISA_QRY = '/content/LISA.QUE'\n",
        "PATH_TO_LISA_REL = '/content/LISARJ.NUM'\n",
        "\n",
        "from collections import defaultdict\n",
        "import re\n",
        "import json\n",
        "from io import StringIO\n",
        "import numpy as np\n",
        "import os\n",
        "\n",
        "# get the text and query files\n",
        "\n",
        "file_regex = re.compile('LISA[0-5]')\n",
        "lisa_files = [i for i in os.listdir(PATH_TO_LISA_TXT) if os.path.isfile(os.path.join(PATH_TO_LISA_TXT,i)) and re.match(file_regex,i)]\n",
        "\n",
        "txt_entry_marker = re.compile('\\*{44}',re.MULTILINE)\n",
        "qry_entry_marker = re.compile('#')\n",
        "\n",
        "def get_data(PATH_TO_FILES, marker):\n",
        "  \"\"\"\n",
        "  Reads multiple files and spilts text into entries at the entry marker.\n",
        "  The 'marker' contains the regex at which we want to split\n",
        "  Pops last element since it's empty.\n",
        "  \"\"\"\n",
        "  with open (PATH_TO_FILES,'r') as f:\n",
        "    text = f.read().replace('     ','')\n",
        "    lines = re.split(marker,text)\n",
        "    lines.pop()\n",
        "  return lines\n",
        "\n",
        "lisa_txt_list = []\n",
        "for name in lisa_files:\n",
        "  lisa_txt_list.extend(get_data(PATH_TO_LISA_TXT+name, txt_entry_marker))\n",
        "\n",
        "lisa_qry_list = get_data(PATH_TO_LISA_QRY, qry_entry_marker)\n",
        "\n",
        "# process text file\n",
        "\n",
        "doc_strip = re.compile('\\n?Document {1,2}')\n",
        "\n",
        "lisa_txt_list_stripped = []\n",
        "lisa_txt_data = defaultdict(dict)\n",
        "\n",
        "for el in lisa_txt_list:\n",
        "  lisa_txt_list_stripped.append(re.sub(doc_strip,'', el))\n",
        "\n",
        "for entry in lisa_txt_list_stripped:\n",
        "  parts = entry.split('\\n')\n",
        "  empty_index = parts.index('')\n",
        "  ID = parts[0].strip(' ')\n",
        "  title = parts[1:empty_index]\n",
        "  text = parts[empty_index+1:]\n",
        "  lisa_txt_data[ID]['title'] = ''.join(title)\n",
        "  lisa_txt_data[ID]['text'] = ''.join(text)\n",
        "\n",
        "# process the query data\n",
        "\n",
        "lisa_qry_rel_data = defaultdict(dict)\n",
        "# first line is a special case because it doesn't start with a newline\n",
        "first_line = lisa_qry_list[0]\n",
        "first_question = first_line[1:]\n",
        "lisa_qry_rel_data[int(first_line[0])]['question'] = first_question\n",
        "\n",
        "# after that every line can be handle the same way\n",
        "for n in range(0,len(lisa_qry_list)-1):\n",
        "  line = re.split('\\n',lisa_qry_list[n+1])\n",
        "  question = ' '.join(line[2:])\n",
        "  lisa_qry_rel_data[int(line[1])]['question'] = question\n",
        "\n",
        "# process relevance assesments\n",
        "lisa_rel = defaultdict(list)\n",
        "\n",
        "rel_marker = re.compile('            \\n        ')\n",
        "rel_split = re.compile('\\n')\n",
        "\n",
        "with open (PATH_TO_LISA_REL,'r') as f:\n",
        "  file = f.read().strip('       ').replace('\\n','')\n",
        "  lines = re.split(' ',file)\n",
        "  lines = list(filter(None, lines))\n",
        "  n = 0\n",
        "  while n < len(lines):\n",
        "    ID = int(lines[n])\n",
        "    num_rel = int(lines[n+1])\n",
        "    rels = lines[(n+2):(n+num_rel+2)]\n",
        "    lisa_rel[ID].extend(rels)\n",
        "    n = n+1+num_rel+1\n",
        "\n",
        "for id, rels in lisa_rel.items():\n",
        "  lisa_qry_rel_data[id]['relevance_assessments'] = [int(r) for r in rels]\n",
        "\n",
        "# transform text to sentences for BERT text search\n",
        "import nltk\n",
        "from nltk import tokenize\n",
        "from nltk.tokenize import sent_tokenize\n",
        "nltk.download('punkt')\n",
        "\n",
        "def text_to_sentences(string):\n",
        "  sentences = tokenize.sent_tokenize(string)\n",
        "  return sentences\n",
        "\n",
        "for ID, doc in lisa_txt_data.items():\n",
        "  text = lisa_txt_data[ID]['text']\n",
        "  lisa_txt_data[ID]['sentences'] = text_to_sentences(text)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-04-29 11:35:05--  http://ir.dcs.gla.ac.uk/resources/test_collections/lisa/lisa.tar.gz\n",
            "Resolving ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)... 130.209.240.253\n",
            "Connecting to ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)|130.209.240.253|:80... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 1227725 (1.2M) [application/gzip]\n",
            "Saving to: ‘lisa.tar.gz’\n",
            "\n",
            "lisa.tar.gz         100%[===================>]   1.17M  1.85MB/s    in 0.6s    \n",
            "\n",
            "2021-04-29 11:35:06 (1.85 MB/s) - ‘lisa.tar.gz’ saved [1227725/1227725]\n",
            "\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AHM9RdKc1qRJ"
      },
      "source": [
        "**Indexing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "MwkcdC8p1qRJ",
        "outputId": "5641d62c-ed34-489c-86b6-d553ef44f8c3"
      },
      "source": [
        "# Elasticsearch settings, without using Opendistro\n",
        "# Every index has title, text and vectors to fill\n",
        "settings_nest = {\n",
        "  \"mappings\": {\n",
        "    \"properties\": {\n",
        "      \"text_vector\": {\n",
        "        \"type\": \"nested\",\n",
        "        \"properties\": {\n",
        "          \"vector\": {\n",
        "            \"type\": \"dense_vector\",\n",
        "            \"dims\": 768\n",
        "          }\n",
        "        }\n",
        "      },\n",
        "     \"text\": {\n",
        "            \"type\" : \"text\"\n",
        "          },\n",
        "          \"title_vector\": {\n",
        "           \"type\": \"dense_vector\",\n",
        "           \"dims\": 768\n",
        "            },\n",
        "      \"title\": {\n",
        "              \"type\": \"text\"\n",
        "              },\n",
        "    }\n",
        "  }\n",
        "}\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "lisa_index = \"lisa-corpus\"\n",
        "es.indices.delete(index=lisa_index, ignore=[400, 404])\n",
        "es.indices.create(lisa_index)\n",
        "\n",
        "lisa_index_sent = \"bert-sentence-lisa-corpus\"\n",
        "es.indices.delete(index=lisa_index_sent, ignore=[400, 404])\n",
        "es.indices.create(lisa_index_sent, body=settings_nest)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True,\n",
              " 'index': 'bert-sentence-lisa-corpus',\n",
              " 'shards_acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 24
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "U7ZwCYEf1qRJ",
        "outputId": "8d9b5b17-cfb9-4d27-8d0d-a09febf2c8bb"
      },
      "source": [
        "for ID, doc_data in lisa_txt_data.items():\n",
        "    # Index for match query\n",
        "    es.index(index=lisa_index, id=ID, body=doc_data)\n",
        "    # Index for BERT search\n",
        "    es.index(\n",
        "      index=lisa_index_sent,\n",
        "      id=ID,\n",
        "      body={\n",
        "          'text_vector': [{\"vector\": model.encode(text)} for text in lisa_txt_data[ID]['sentences']],\n",
        "          'text': lisa_txt_data[ID]['text'],\n",
        "          'title_vector': model.encode(''.join(lisa_txt_data[ID]['title'])),\n",
        "          'title': lisa_txt_data[ID]['title']\n",
        "          }\n",
        "    )\n",
        "#print new index list\n",
        "create_response = es.cat.indices()\n",
        "print(create_response)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "yellow open bert-sentence-cacm-corpus      wQ-lRS5ISYm1FT4vzZHnNQ 1 1 13353 0 191.4mb 191.4mb\n",
            "yellow open bert-sentence-lisa-corpus      nq5XvEIcTOigd0CwfEuUCA 1 1   953 0 325.4mb 325.4mb\n",
            "yellow open cranfield-corpus               j5RF5ISFSfyqm3hE-_TCBw 1 1  1400 0   4.5mb   4.5mb\n",
            "yellow open bert-sentence-cranfield-corpus _VSu6MP3ToagtcTR3APXXw 1 1 11515 0 171.9mb 171.9mb\n",
            "yellow open cacm-corpus                    hW80LSBXRtalOg57jYJoJw 1 1  3204 0   4.9mb   4.9mb\n",
            "yellow open cisi-corpus                    bQuJ-NwwTmC1xWgiEQFmrw 1 1  1460 0   4.4mb   4.4mb\n",
            "yellow open adi-corpus                     VhHKO45mTSuy4SRdtN199Q 1 1    82 0 160.5kb 160.5kb\n",
            "yellow open lisa-corpus                    vgXDjpBjSqCyLvhmnaWToQ 1 1   200 0   2.8mb   2.8mb\n",
            "yellow open bert-sentence-cisi-corpus      xCRbxtYUSEGFZ8f1e4nmOg 1 1  7792 0 116.7mb 116.7mb\n",
            "yellow open bert-sentence-adi-corpus       zX-Lk5SnQXOVzBcpw5IusQ 1 1   285 0   4.3mb   4.3mb\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "as_Xf08e1qRK"
      },
      "source": [
        "### Medline Corpus"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NbFhHCto1qRK"
      },
      "source": [
        "You can get the corpus from [this link](http://ir.dcs.gla.ac.uk/resources/test_collections/med/).  <br>\n",
        "For detailed information about the format of the files, see the PragmaLingu [ Benchmarks](https://pragmalingu.de/docs/benchmarks/overview)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "N2p2U2T61qRK"
      },
      "source": [
        "**Parsing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qJWGdOx11qRK",
        "outputId": "fab41280-ebea-4463-86cd-584f13b7d47b"
      },
      "source": [
        "# download and unzip data\n",
        "!wget http://ir.dcs.gla.ac.uk/resources/test_collections/medl/med.tar.gz\n",
        "!tar -xf med.tar.gz\n",
        "\n",
        "# set paths to the dowloaded data as variables\n",
        "PATH_TO_MED_TXT = '/content/MED.ALL'\n",
        "PATH_TO_MED_QRY = '/content/MED.QRY'\n",
        "PATH_TO_MED_REL = '/content/MED.REL'\n",
        "\n",
        "from collections import defaultdict\n",
        "import re\n",
        "import json\n",
        "from io import StringIO\n",
        "import numpy as np\n",
        "\n",
        "\n",
        "# get the text and query files\n",
        "\n",
        "ID_marker = re.compile('\\.I')\n",
        "\n",
        "def get_data(PATH_TO_FILE, marker):\n",
        "  \"\"\"\n",
        "  Reads file and spilts text into entries at the ID marker '.I'.\n",
        "  First entry is empty, so it's removed.\n",
        "  'marker' contains the regex at which we want to split\n",
        "  \"\"\"\n",
        "  with open (PATH_TO_FILE,'r') as f:\n",
        "    text = f.read().replace('\\n',\" \")\n",
        "    lines = re.split(marker,text)\n",
        "    lines.pop(0)\n",
        "  return lines\n",
        "\n",
        "med_txt_list = get_data(PATH_TO_MED_TXT, ID_marker)\n",
        "med_qry_list = get_data(PATH_TO_MED_QRY, ID_marker)\n",
        "\n",
        "# process the text and query files\n",
        "\n",
        "import nltk\n",
        "from nltk import tokenize\n",
        "from nltk.tokenize import sent_tokenize\n",
        "nltk.download('punkt')\n",
        "\n",
        "def text_to_sentences(string):\n",
        "  sentences = tokenize.sent_tokenize(string)\n",
        "  return sentences\n",
        "\n",
        "qry_chunk_start = re.compile('\\.W')\n",
        "\n",
        "med_txt_data = defaultdict(dict)\n",
        "med_qry_rel_data = defaultdict(dict)\n",
        "\n",
        "def fill_dictionary(dictionary, chunk_list, marker, key_name):\n",
        "  for n in range(0,len(chunk_list)):\n",
        "    line = chunk_list[n]\n",
        "    _ , chunk = re.split(marker,line)\n",
        "    dictionary[n+1][key_name] = chunk.strip()\n",
        "\n",
        "fill_dictionary(med_txt_data, med_txt_list, qry_chunk_start, 'text')\n",
        "fill_dictionary(med_qry_rel_data, med_qry_list, qry_chunk_start, 'question')\n",
        "\n",
        "# process relevance assesments without rating\n",
        "med_rel = defaultdict(list)\n",
        "\n",
        "with open (PATH_TO_MED_REL,'r') as f:\n",
        "  for line in f:\n",
        "    line = re.split(' ',line)\n",
        "    med_rel[int(line[0])].append(line[2])\n",
        "\n",
        "for id, rels in med_rel.items():\n",
        "  med_qry_rel_data[id]['relevance_assessments'] = [int(r) for r in rels]\n",
        "\n",
        "\n",
        "# transform text to sentences for BERT text search\n",
        "import nltk\n",
        "from nltk import tokenize\n",
        "from nltk.tokenize import sent_tokenize\n",
        "nltk.download('punkt')\n",
        "\n",
        "def text_to_sentences(string):\n",
        "  sentences = tokenize.sent_tokenize(string)\n",
        "  return sentences\n",
        "\n",
        "for ID, doc in med_txt_data.items():\n",
        "  text = med_txt_data[ID]['text']\n",
        "  med_txt_data[ID]['sentences'] = text_to_sentences(text)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-04-29 11:46:38--  http://ir.dcs.gla.ac.uk/resources/test_collections/medl/med.tar.gz\n",
            "Resolving ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)... 130.209.240.253\n",
            "Connecting to ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)|130.209.240.253|:80... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 375183 (366K) [application/gzip]\n",
            "Saving to: ‘med.tar.gz’\n",
            "\n",
            "med.tar.gz          100%[===================>] 366.39K   851KB/s    in 0.4s    \n",
            "\n",
            "2021-04-29 11:46:39 (851 KB/s) - ‘med.tar.gz’ saved [375183/375183]\n",
            "\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rbimvEOn1qRL"
      },
      "source": [
        "**Indexing**\n",
        "\n",
        "We won't index the title BERT search because Medline, NPL, and Time don't have titles."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-9A6SEkz1qRL",
        "outputId": "0ac4da4c-47a3-46d8-9d7a-1ca14bda4bf8"
      },
      "source": [
        "# Elasticsearch settings, without using Opendistro\n",
        "# Every index has title, text and vectors to fill\n",
        "settings_nest = {\n",
        "  \"mappings\": {\n",
        "    \"properties\": {\n",
        "      \"text_vector\": {\n",
        "        \"type\": \"nested\",\n",
        "        \"properties\": {\n",
        "          \"vector\": {\n",
        "            \"type\": \"dense_vector\",\n",
        "            \"dims\": 768\n",
        "          }\n",
        "        }\n",
        "      },\n",
        "     \"text\": {\n",
        "            \"type\" : \"text\"\n",
        "          },\n",
        "          \"title_vector\": {\n",
        "           \"type\": \"dense_vector\",\n",
        "           \"dims\": 768\n",
        "            },\n",
        "      \"title\": {\n",
        "              \"type\": \"text\"\n",
        "              },\n",
        "    }\n",
        "  }\n",
        "}\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "med_index = \"medline-corpus\"\n",
        "es.indices.delete(index=med_index, ignore=[400, 404])\n",
        "es.indices.create(med_index)\n",
        "\n",
        "med_index_sent = \"bert-sentence-medline-corpus\"\n",
        "es.indices.delete(index=med_index_sent, ignore=[400, 404])\n",
        "es.indices.create(med_index_sent, body=settings_nest)\n"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True,\n",
              " 'index': 'bert-sentence-medline-corpus',\n",
              " 'shards_acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "__o7GqKr1qRL",
        "outputId": "fd482e7e-4eda-4a55-e76c-a2039559f20f"
      },
      "source": [
        "for ID, doc_data in med_txt_data.items():\n",
        "  # Index for match query\n",
        "  es.index(index=med_index, id=ID, body=doc_data)\n",
        "  # Index for BERT search\n",
        "  es.index(\n",
        "      index=med_index_sent,\n",
        "      id=ID,\n",
        "      body={\n",
        "          'text_vector': [{\"vector\": model.encode(text)} for text in med_txt_data[ID]['sentences']],\n",
        "          'text': med_txt_data[ID]['text']\n",
        "          }\n",
        "    )\n",
        "\n",
        "#print new index list\n",
        "create_response = es.cat.indices()\n",
        "print(create_response)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "yellow open bert-sentence-cacm-corpus      wQ-lRS5ISYm1FT4vzZHnNQ 1 1 13353 0 191.4mb 191.4mb\n",
            "yellow open bert-sentence-lisa-corpus      nq5XvEIcTOigd0CwfEuUCA 1 1 25833 0   400mb   400mb\n",
            "yellow open cranfield-corpus               j5RF5ISFSfyqm3hE-_TCBw 1 1  1400 0   4.5mb   4.5mb\n",
            "yellow open bert-sentence-cranfield-corpus _VSu6MP3ToagtcTR3APXXw 1 1 11515 0 171.9mb 171.9mb\n",
            "yellow open cacm-corpus                    hW80LSBXRtalOg57jYJoJw 1 1  3204 0   4.9mb   4.9mb\n",
            "yellow open cisi-corpus                    bQuJ-NwwTmC1xWgiEQFmrw 1 1  1460 0   4.4mb   4.4mb\n",
            "yellow open adi-corpus                     VhHKO45mTSuy4SRdtN199Q 1 1    82 0 160.5kb 160.5kb\n",
            "yellow open medline-corpus                 _0NBMYlNRT6b5WagtUrQyg 1 1   157 0   1.6mb   1.6mb\n",
            "yellow open lisa-corpus                    vgXDjpBjSqCyLvhmnaWToQ 1 1   200 0   2.8mb   2.8mb\n",
            "yellow open bert-sentence-medline-corpus   _mAnt5d0Qf28xt606nJnZQ 1 1  1319 0   107mb   107mb\n",
            "yellow open bert-sentence-cisi-corpus      xCRbxtYUSEGFZ8f1e4nmOg 1 1  7792 0 116.7mb 116.7mb\n",
            "yellow open bert-sentence-adi-corpus       zX-Lk5SnQXOVzBcpw5IusQ 1 1   285 0   4.3mb   4.3mb\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "x9TyBrSC1qRM"
      },
      "source": [
        "### NPL Corpus"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lRPenLco1qRM"
      },
      "source": [
        "You can get the corpus from [this link](http://ir.dcs.gla.ac.uk/resources/test_collections/npl/).  <br>\n",
        "For detailed information about the format of the files, see the PragmaLingu [ Benchmarks](https://pragmalingu.de/docs/benchmarks/overview)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "M1a_LU-U1qRM"
      },
      "source": [
        "**Parsing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "v1dIV-2W1qRM",
        "outputId": "ef2ee2e9-aa21-4d74-deca-cc0078bb8562"
      },
      "source": [
        "# download and unzip data\n",
        "!wget http://ir.dcs.gla.ac.uk/resources/test_collections/npl/npl.tar.gz\n",
        "!tar -xf npl.tar.gz\n",
        "\n",
        "# set paths to the dowloaded data as variablesDownload and unzip data.\n",
        "\n",
        "PATH_TO_NPL_TXT = '/content/doc-text'\n",
        "PATH_TO_NPL_QRY = '/content/query-text'\n",
        "PATH_TO_NPL_REL = '/content/rlv-ass'\n",
        "\n",
        "from collections import defaultdict\n",
        "import re\n",
        "import json\n",
        "from io import StringIO\n",
        "import numpy as np\n",
        "import os\n",
        "\n",
        "npl_index = \"npl-corpus\"\n",
        "\n",
        "# get the text, query and rel files\n",
        "\n",
        "txt_entry_marker = re.compile('\\n   /\\n')\n",
        "qry_entry_marker = re.compile('\\n/\\n')\n",
        "rel_entry_marker = re.compile('\\n   /\\n')\n",
        "\n",
        "def get_data(PATH_TO_FILES, marker):\n",
        "  \"\"\"\n",
        "  Reads multiple files and spilts text into entries at the entry marker.\n",
        "  The 'marker' contains the regex at which we want to split\n",
        "  Pops last element since it's empty.\n",
        "  \"\"\"\n",
        "  with open (PATH_TO_FILES,'r') as f:\n",
        "    text = f.read()\n",
        "    lines = re.split(marker,text)\n",
        "    lines.pop()\n",
        "  return lines\n",
        "\n",
        "npl_txt_list = get_data(PATH_TO_NPL_TXT, txt_entry_marker)\n",
        "npl_qry_list = get_data(PATH_TO_NPL_QRY, qry_entry_marker)\n",
        "npl_rel_list = get_data(PATH_TO_NPL_REL, rel_entry_marker)\n",
        "\n",
        "# process the documents\n",
        "\n",
        "npl_txt_data = defaultdict(dict)\n",
        "\n",
        "for entry in npl_txt_list:\n",
        "  splitted = entry.split('\\n')\n",
        "  splitted = list(filter(None, splitted))\n",
        "  ID = splitted[0]\n",
        "  text = ' '.join(map(str, splitted[1:]))\n",
        "  npl_txt_data[ID]['text'] = text\n",
        "\n",
        "# process the queries\n",
        "\n",
        "npl_qry_rel_data = defaultdict(dict)\n",
        "\n",
        "for entry in npl_qry_list:\n",
        "  splitted = entry.split('\\n')\n",
        "  splitted = list(filter(None, splitted))\n",
        "  ID = int(splitted[0])\n",
        "  question = ' '.join(map(str, splitted[1:]))\n",
        "  npl_qry_rel_data[ID]['question'] = question\n",
        "\n",
        "# process relevance assesments\n",
        "\n",
        "npl_rel = defaultdict(list)\n",
        "\n",
        "for entry in npl_rel_list:\n",
        "  splitted = entry.replace('\\n','').split(' ')\n",
        "  splitted = list(filter(None, splitted))\n",
        "  ID = int(splitted[0])\n",
        "  npl_rel[ID].extend(splitted[1:])\n",
        "\n",
        "npl_rel[60] = npl_rel[6010182]\n",
        "npl_rel[60].append('10182')\n",
        "del npl_rel[6010182]\n",
        "\n",
        "for id, rels in npl_rel.items():\n",
        "  npl_qry_rel_data[id]['relevance_assessments'] = [int(r) for r in rels]\n",
        "\n",
        "# transform text to sentences for BERT text search\n",
        "import nltk\n",
        "from nltk import tokenize\n",
        "from nltk.tokenize import sent_tokenize\n",
        "nltk.download('punkt')\n",
        "\n",
        "def text_to_sentences(string):\n",
        "  sentences = tokenize.sent_tokenize(string)\n",
        "  return sentences\n",
        "\n",
        "for ID, doc in npl_txt_data.items():\n",
        "  text = npl_txt_data[ID]['text']\n",
        "  npl_txt_data[ID]['sentences'] = text_to_sentences(text)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-04-29 11:49:43--  http://ir.dcs.gla.ac.uk/resources/test_collections/npl/npl.tar.gz\n",
            "Resolving ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)... 130.209.240.253\n",
            "Connecting to ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)|130.209.240.253|:80... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 2125168 (2.0M) [application/gzip]\n",
            "Saving to: ‘npl.tar.gz’\n",
            "\n",
            "npl.tar.gz          100%[===================>]   2.03M  2.73MB/s    in 0.7s    \n",
            "\n",
            "2021-04-29 11:49:44 (2.73 MB/s) - ‘npl.tar.gz’ saved [2125168/2125168]\n",
            "\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "x6NzqSO_1qRN"
      },
      "source": [
        "**Indexing**\n",
        "\n",
        "We won't index the title BERT search because Medline, NPL, and Time don't have titles."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "vozK3w0l1qRN",
        "outputId": "de63fac7-945e-4fb8-dc7a-18988e9073a5"
      },
      "source": [
        "# Elasticsearch settings, without using Opendistro\n",
        "# Every index has title, text and vectors to fill\n",
        "settings_nest = {\n",
        "  \"mappings\": {\n",
        "    \"properties\": {\n",
        "      \"text_vector\": {\n",
        "        \"type\": \"nested\",\n",
        "        \"properties\": {\n",
        "          \"vector\": {\n",
        "            \"type\": \"dense_vector\",\n",
        "            \"dims\": 768\n",
        "          }\n",
        "        }\n",
        "      },\n",
        "     \"text\": {\n",
        "            \"type\" : \"text\",\n",
        "          },\n",
        "          \"title\": {\n",
        "              \"type\": \"text\"\n",
        "              },\n",
        "          \"title_vector\": {\n",
        "           \"type\": \"dense_vector\",\n",
        "           \"dims\": 768\n",
        "            }\n",
        "    }\n",
        "  }\n",
        "}\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "\n",
        "npl_index = \"npl-corpus\"\n",
        "es.indices.delete(index=npl_index, ignore=[400, 404])\n",
        "es.indices.create(npl_index)\n",
        "\n",
        "npl_index_sent = \"bert-sentence-npl-corpus\"\n",
        "es.indices.delete(index=npl_index_sent, ignore=[400, 404])\n",
        "es.indices.create(npl_index_sent, body=settings_nest)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True,\n",
              " 'index': 'bert-sentence-npl-corpus',\n",
              " 'shards_acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 30
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lzwyONbo1qRO",
        "outputId": "c4ddff73-6548-45ea-fe0d-46075c9b1c43"
      },
      "source": [
        "for ID, doc_data in npl_txt_data.items():\n",
        "  # Index for match query\n",
        "  es.index(index=npl_index, id=ID, body=doc_data)\n",
        "  # Index for BERT search\n",
        "  text_list = []\n",
        "  es.index(\n",
        "      index=npl_index_sent,\n",
        "      id=ID,\n",
        "      body={\n",
        "          'text_vector': [{\"vector\": model.encode(text)} for text in npl_txt_data[ID]['sentences']],\n",
        "          'text': npl_txt_data[ID]['text']\n",
        "          }\n",
        "    )\n",
        "\n",
        "#print new index list\n",
        "create_response = es.cat.indices()\n",
        "print(create_response)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "yellow open bert-sentence-lisa-corpus      nq5XvEIcTOigd0CwfEuUCA 1 1 25833 0 385.9mb 385.9mb\n",
            "yellow open cranfield-corpus               j5RF5ISFSfyqm3hE-_TCBw 1 1  1400 0   4.5mb   4.5mb\n",
            "yellow open bert-sentence-cranfield-corpus _VSu6MP3ToagtcTR3APXXw 1 1 11515 0 171.9mb 171.9mb\n",
            "yellow open adi-corpus                     VhHKO45mTSuy4SRdtN199Q 1 1    82 0 160.5kb 160.5kb\n",
            "yellow open medline-corpus                 _0NBMYlNRT6b5WagtUrQyg 1 1  1033 0   3.2mb   3.2mb\n",
            "yellow open lisa-corpus                    vgXDjpBjSqCyLvhmnaWToQ 1 1  6000 0  10.1mb  10.1mb\n",
            "yellow open bert-sentence-adi-corpus       zX-Lk5SnQXOVzBcpw5IusQ 1 1   285 0   4.3mb   4.3mb\n",
            "yellow open bert-sentence-medline-corpus   _mAnt5d0Qf28xt606nJnZQ 1 1  8292 0 108.5mb 108.5mb\n",
            "yellow open bert-sentence-cacm-corpus      wQ-lRS5ISYm1FT4vzZHnNQ 1 1 13353 0 191.4mb 191.4mb\n",
            "yellow open npl-corpus                     p-h-_EmsQK-nGzlbpwj5eA 1 1   610 0   2.9mb   2.9mb\n",
            "yellow open cacm-corpus                    hW80LSBXRtalOg57jYJoJw 1 1  3204 0   4.9mb   4.9mb\n",
            "yellow open cisi-corpus                    bQuJ-NwwTmC1xWgiEQFmrw 1 1  1460 0   4.4mb   4.4mb\n",
            "yellow open bert-sentence-npl-corpus       cBNtDHZnRLGG317PQ0cS5g 1 1  1230 0 139.4mb 139.4mb\n",
            "yellow open bert-sentence-cisi-corpus      xCRbxtYUSEGFZ8f1e4nmOg 1 1  7792 0 116.7mb 116.7mb\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BqYVCeXG1qRO"
      },
      "source": [
        "### Time Corpus\n",
        "\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yhgHdTpP1qRO"
      },
      "source": [
        "You can get the corpus from [this link](http://ir.dcs.gla.ac.uk/resources/test_collections/time/).  <br>\n",
        "For detailed information about the format of the files, see the PragmaLingu [ Benchmarks](https://pragmalingu.de/docs/benchmarks/overview)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ka0MnFZo1qRO"
      },
      "source": [
        "**Parsing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Dy2nr9NN1qRO",
        "outputId": "f1a1f1f8-6f85-417a-f225-79c8fe538610"
      },
      "source": [
        "# download and unzip data\n",
        "!wget http://ir.dcs.gla.ac.uk/resources/test_collections/time/time.tar.gz\n",
        "!tar -xf time.tar.gz\n",
        "\n",
        "# set paths to the dowloaded data as variablesDownload and unzip data.\n",
        "\n",
        "PATH_TO_TIME_TXT = '/content/TIME.ALL'\n",
        "PATH_TO_TIME_QRY = '/content/TIME.QUE'\n",
        "PATH_TO_TIME_REL = '/content/TIME.REL'\n",
        "\n",
        "from collections import defaultdict\n",
        "import re\n",
        "import json\n",
        "from io import StringIO\n",
        "import numpy as np\n",
        "import os\n",
        "\n",
        "time_index = \"time-corpus\"\n",
        "\n",
        "# get the text and query files\n",
        "\n",
        "txt_entry_marker = re.compile('\\*TEXT')\n",
        "qry_entry_marker = re.compile('\\*FIND    ')\n",
        "def get_data(PATH_TO_FILES, marker):\n",
        "  \"\"\"\n",
        "  Reads multiple files and spilts text into entries at the entry marker.\n",
        "  The 'marker' contains the regex at which we want to split\n",
        "  Pops last element since it's empty.\n",
        "  \"\"\"\n",
        "  with open (PATH_TO_FILES,'r') as f:\n",
        "    text = f.read()\n",
        "    lines = re.split(marker,text)\n",
        "    lines.pop(0)\n",
        "  return lines\n",
        "\n",
        "time_txt_list = get_data(PATH_TO_TIME_TXT, txt_entry_marker)\n",
        "time_qry_list = get_data(PATH_TO_TIME_QRY, qry_entry_marker)\n",
        "\n",
        "# process text file\n",
        "page_split = re.compile('PAGE \\d{3}')\n",
        "\n",
        "time_txt_data = defaultdict(dict)\n",
        "ID = 1\n",
        "for entry in time_txt_list:\n",
        "  splitted = re.split(page_split,entry)\n",
        "  time_txt_data[ID]['text'] = splitted[1]\n",
        "  ID += 1\n",
        "\n",
        "\n",
        "# process query file\n",
        "time_qry_rel_data = defaultdict(dict)\n",
        "qry_split = re.compile('\\n\\n ')\n",
        "\n",
        "for query in time_qry_list:\n",
        "  entries = re.split(qry_split,query)\n",
        "  time_qry_rel_data[int(entries[0])]['question'] = entries[1]\n",
        "\n",
        "# process relevance assesments\n",
        "time_rel = defaultdict(list)\n",
        "\n",
        "rel_marker = re.compile('            \\n        ')\n",
        "rel_split = re.compile('\\n')\n",
        "\n",
        "with open (PATH_TO_TIME_REL,'r') as f:\n",
        "  for lines in f:\n",
        "    line = lines.strip().replace('   ',' ').replace('  ',' ').split(' ')\n",
        "    if len(line) > 1:\n",
        "      time_rel[int(line[0])].extend(line[1:])\n",
        "\n",
        "for id, rels in time_rel.items():\n",
        "  time_qry_rel_data[id]['relevance_assessments'] = [int(r) for r in rels]\n",
        "\n",
        "# transform text to sentences for BERT text search\n",
        "import nltk\n",
        "from nltk import tokenize\n",
        "from nltk.tokenize import sent_tokenize\n",
        "nltk.download('punkt')\n",
        "\n",
        "def text_to_sentences(string):\n",
        "  sentences = tokenize.sent_tokenize(string)\n",
        "  return sentences\n",
        "\n",
        "for ID, doc in time_txt_data.items():\n",
        "  text = time_txt_data[ID]['text']\n",
        "  time_txt_data[ID]['sentences'] = text_to_sentences(text)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2021-04-29 11:59:02--  http://ir.dcs.gla.ac.uk/resources/test_collections/time/time.tar.gz\n",
            "Resolving ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)... 130.209.240.253\n",
            "Connecting to ir.dcs.gla.ac.uk (ir.dcs.gla.ac.uk)|130.209.240.253|:80... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 627965 (613K) [application/gzip]\n",
            "Saving to: ‘time.tar.gz’\n",
            "\n",
            "time.tar.gz         100%[===================>] 613.25K  1.13MB/s    in 0.5s    \n",
            "\n",
            "2021-04-29 11:59:03 (1.13 MB/s) - ‘time.tar.gz’ saved [627965/627965]\n",
            "\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "f-zaPFlS1qRP"
      },
      "source": [
        "**Indexing**\n",
        "\n",
        "We won't index the title BERT search because Medline, NPL, and Time don't have titles."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "FZ1giXmw1qRP",
        "outputId": "70936691-a7e6-48e6-813a-5af19770354e"
      },
      "source": [
        "# Elasticsearch settings, without using Opendistro\n",
        "# Every index has title, text and vectors to fill\n",
        "settings_nest = {\n",
        "  \"mappings\": {\n",
        "    \"properties\": {\n",
        "      \"text_vector\": {\n",
        "        \"type\": \"nested\",\n",
        "        \"properties\": {\n",
        "          \"vector\": {\n",
        "            \"type\": \"dense_vector\",\n",
        "            \"dims\": 768\n",
        "          }\n",
        "        }\n",
        "      },\n",
        "     \"text\": {\n",
        "            \"type\" : \"text\",\n",
        "          },\n",
        "          \"title\": {\n",
        "              \"type\": \"text\"\n",
        "              },\n",
        "          \"title_vector\": {\n",
        "           \"type\": \"dense_vector\",\n",
        "           \"dims\": 768\n",
        "            }\n",
        "    }\n",
        "  }\n",
        "}\n",
        "\n",
        "#create index, see https://elasticsearch-py.readthedocs.io/en/master/api.html#elasticsearch.client.IndicesClient.create\n",
        "\n",
        "time_index = \"time-corpus\"\n",
        "es.indices.delete(index=time_index, ignore=[400, 404])\n",
        "es.indices.create(time_index)\n",
        "\n",
        "time_index_sent = \"bert-sentence-time-corpus\"\n",
        "es.indices.delete(index=time_index_sent, ignore=[400, 404])\n",
        "es.indices.create(time_index_sent, body=settings_nest)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'acknowledged': True,\n",
              " 'index': 'bert-sentence-time-corpus',\n",
              " 'shards_acknowledged': True}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 33
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "y69mNJyu1qRQ",
        "outputId": "f48db2f1-c4b7-4a71-f3fd-23ad1eefd3dc"
      },
      "source": [
        "for ID, doc_data in time_txt_data.items():\n",
        "  # Index for match query\n",
        "  es.index(index=time_index, id=ID, body=doc_data)\n",
        "  # Index for BERT search\n",
        "  es.index(\n",
        "      index=time_index_sent,\n",
        "      id=ID,\n",
        "      body={\n",
        "          'text_vector': [{\"vector\": model.encode(text)} for text in time_txt_data[ID]['sentences']],\n",
        "          'text': time_txt_data[ID]['text']\n",
        "          }\n",
        "    )\n",
        "\n",
        "#print new index list\n",
        "create_response = es.cat.indices()\n",
        "print(create_response)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "yellow open bert-sentence-lisa-corpus      nq5XvEIcTOigd0CwfEuUCA 1 1 25833 0 385.9mb 385.9mb\n",
            "yellow open cranfield-corpus               j5RF5ISFSfyqm3hE-_TCBw 1 1  1400 0   4.5mb   4.5mb\n",
            "yellow open bert-sentence-cranfield-corpus _VSu6MP3ToagtcTR3APXXw 1 1 11515 0 171.9mb 171.9mb\n",
            "yellow open adi-corpus                     VhHKO45mTSuy4SRdtN199Q 1 1    82 0 160.5kb 160.5kb\n",
            "yellow open medline-corpus                 _0NBMYlNRT6b5WagtUrQyg 1 1  1033 0   3.2mb   3.2mb\n",
            "yellow open lisa-corpus                    vgXDjpBjSqCyLvhmnaWToQ 1 1  6000 0  10.1mb  10.1mb\n",
            "yellow open bert-sentence-adi-corpus       zX-Lk5SnQXOVzBcpw5IusQ 1 1   285 0   4.3mb   4.3mb\n",
            "yellow open bert-sentence-medline-corpus   _mAnt5d0Qf28xt606nJnZQ 1 1  8292 0 108.5mb 108.5mb\n",
            "yellow open bert-sentence-cacm-corpus      wQ-lRS5ISYm1FT4vzZHnNQ 1 1 13353 0 191.4mb 191.4mb\n",
            "yellow open npl-corpus                     p-h-_EmsQK-nGzlbpwj5eA 1 1   610 0   2.9mb   2.9mb\n",
            "yellow open cacm-corpus                    hW80LSBXRtalOg57jYJoJw 1 1  3204 0   4.9mb   4.9mb\n",
            "yellow open cisi-corpus                    bQuJ-NwwTmC1xWgiEQFmrw 1 1  1460 0   4.4mb   4.4mb\n",
            "yellow open bert-sentence-npl-corpus       cBNtDHZnRLGG317PQ0cS5g 1 1  1230 0 141.4mb 141.4mb\n",
            "yellow open time-corpus                    jQijUkfLQFuFbTwhtOOaRw 1 1    47 0   1.9mb   1.9mb\n",
            "yellow open bert-sentence-time-corpus      x_Fj5UbXS8Of2CrbvNUoew 1 1  1245 0 155.8mb 155.8mb\n",
            "yellow open bert-sentence-cisi-corpus      xCRbxtYUSEGFZ8f1e4nmOg 1 1  7792 0 116.7mb 116.7mb\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4IbavyCL1qRQ"
      },
      "source": [
        "## Evaluation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "OHTs6Cn01qRQ"
      },
      "source": [
        "### Recall\n",
        "\n",
        "In this section, we only evaluate the Recall scores."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "uGt0flee1qRQ"
      },
      "source": [
        "**Multi Match Query**\n",
        "\n",
        "Here, we evaluate the data with the [\"multi_match\"](https://pragmalingu.de/docs/experiments/experiment1#standard-elasticsearch) option of Elasticsearch:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RVjF4ztu1qRQ"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/query-dsl-simple-query-string-query.html\n",
        "\n",
        "import json\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'adi-corpus'\n",
        "cacm_index = 'cacm-corpus'\n",
        "cisi_index = 'cisi-corpus'\n",
        "cran_index = 'cranfield-corpus'\n",
        "lisa_index = 'lisa-corpus'\n",
        "med_index = 'medline-corpus'\n",
        "npl_index = 'npl-corpus'\n",
        "time_index = 'time-corpus'\n",
        "\n",
        "text_field = ['text']\n",
        "title_field = ['title']\n",
        "text_title_field = ['title','text']\n",
        "\n",
        "#function to get normal match evaluation body\n",
        "def create_query_body_match_recall(query_rel_dict, rel_dict, index_name,field):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\": '',\n",
        "      \"metric\": {\n",
        "          \"recall\": {\n",
        "              \"relevant_rating_threshold\": 1,\n",
        "              \"k\": 20\n",
        "              }\n",
        "      }\n",
        "  }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\"query\": { \"multi_match\": { \"query\": '' , \"fields\" : field}}}\n",
        "    current_query[\"query\"][\"multi_match\"][\"query\"] = query_txt['question']\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\"_index\": index_name, \"_id\": str(el), \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "\n",
        "#ADI\n",
        "adi_create_match_recall = create_query_body_match_recall(adi_qry_rel_data, adi_rel, adi_index,title_field)\n",
        "adi_eval_body_match_recall = json.dumps(adi_create_match_recall)\n",
        "adi_res_match_recall_title = es.rank_eval(adi_eval_body_match_recall, adi_index)\n",
        "#print(json.dumps(adi_res_match_recall, indent=4, sort_keys=True))\n",
        "adi_create_match_recall = create_query_body_match_recall(adi_qry_rel_data, adi_rel, adi_index,text_field)\n",
        "adi_eval_body_match_recall = json.dumps(adi_create_match_recall)\n",
        "adi_res_match_recall_text = es.rank_eval(adi_eval_body_match_recall, adi_index)\n",
        "#print(json.dumps(adi_create_match_recall, indent=4, sort_keys=True))\n",
        "adi_create_match_recall = create_query_body_match_recall(adi_qry_rel_data, adi_rel, adi_index,text_title_field)\n",
        "adi_eval_body_match_recall = json.dumps(adi_create_match_recall)\n",
        "adi_res_match_recall_text_title = es.rank_eval(adi_eval_body_match_recall, adi_index)\n",
        "#print(json.dumps(adi_create_match_recall, indent=4, sort_keys=True))\n",
        "\n",
        "#CACM\n",
        "cacm_create_match_recall = create_query_body_match_recall(cacm_qry_rel_data, cacm_rel, cacm_index,title_field)\n",
        "cacm_eval_body_match_recall = json.dumps(cacm_create_match_recall)\n",
        "cacm_res_match_recall_title = es.rank_eval(cacm_eval_body_match_recall,cacm_index)\n",
        "#print(json.dumps(cacm_res_match_recall, indent=4, sort_keys=True))\n",
        "cacm_create_match_recall = create_query_body_match_recall(cacm_qry_rel_data, cacm_rel, cacm_index,text_field)\n",
        "cacm_eval_body_match_recall = json.dumps(cacm_create_match_recall)\n",
        "cacm_res_match_recall_text = es.rank_eval(cacm_eval_body_match_recall,cacm_index)\n",
        "#print(json.dumps(cacm_res_match_recall, indent=4, sort_keys=True))\n",
        "cacm_create_match_recall = create_query_body_match_recall(cacm_qry_rel_data, cacm_rel, cacm_index,text_title_field)\n",
        "cacm_eval_body_match_recall = json.dumps(cacm_create_match_recall)\n",
        "cacm_res_match_recall_text_title = es.rank_eval(cacm_eval_body_match_recall,cacm_index)\n",
        "#print(json.dumps(cacm_res_match_recall, indent=4, sort_keys=True))\n",
        "\n",
        "#CISI\n",
        "cisi_create_match_recall = create_query_body_match_recall(cisi_qry_rel_data, cisi_rel, cisi_index,title_field)\n",
        "cisi_eval_body_match_recall = json.dumps(cisi_create_match_recall)\n",
        "cisi_res_match_recall_title = es.rank_eval(cisi_eval_body_match_recall, cisi_index)\n",
        "#print(json.dumps(cisi_res_match_recall, indent=4, sort_keys=True))\n",
        "cisi_create_match_recall = create_query_body_match_recall(cisi_qry_rel_data, cisi_rel, cisi_index,text_field)\n",
        "cisi_eval_body_match_recall = json.dumps(cisi_create_match_recall)\n",
        "cisi_res_match_recall_text = es.rank_eval(cisi_eval_body_match_recall, cisi_index)\n",
        "#print(json.dumps(cisi_res_match_recall, indent=4, sort_keys=True))\n",
        "cisi_create_match_recall = create_query_body_match_recall(cisi_qry_rel_data, cisi_rel, cisi_index,text_title_field)\n",
        "cisi_eval_body_match_recall = json.dumps(cisi_create_match_recall)\n",
        "cisi_res_match_recall_text_title = es.rank_eval(cisi_eval_body_match_recall, cisi_index)\n",
        "#print(json.dumps(cisi_res_match_recall, indent=4, sort_keys=True))\n",
        "\n",
        "#Cranfield\n",
        "cran_create_match_recall = create_query_body_match_recall(cran_qry_rel_data, cran_rel, cran_index,title_field)\n",
        "cran_eval_body_match_recall = json.dumps(cran_create_match_recall)\n",
        "cran_res_match_recall_title = es.rank_eval(cran_eval_body_match_recall, cran_index)\n",
        "#print(json.dumps(cran_create_match_recall, indent=4, sort_keys=True))\n",
        "cran_create_match_recall = create_query_body_match_recall(cran_qry_rel_data, cran_rel, cran_index,text_field)\n",
        "cran_eval_body_match_recall = json.dumps(cran_create_match_recall)\n",
        "cran_res_match_recall_text = es.rank_eval(cran_eval_body_match_recall, cran_index)\n",
        "#print(json.dumps(cran_create_match_recall, indent=4, sort_keys=True))\n",
        "cran_create_match_recall = create_query_body_match_recall(cran_qry_rel_data, cran_rel, cran_index,text_title_field)\n",
        "cran_eval_body_match_recall = json.dumps(cran_create_match_recall)\n",
        "cran_res_match_recall_text_title = es.rank_eval(cran_eval_body_match_recall, cran_index)\n",
        "#print(json.dumps(cran_create_match_recall, indent=4, sort_keys=True))\n",
        "\n",
        "#LISA\n",
        "lisa_create_match_recall = create_query_body_match_recall(lisa_qry_rel_data, lisa_rel, lisa_index,title_field)\n",
        "lisa_eval_body_match_recall = json.dumps(lisa_create_match_recall)\n",
        "lisa_res_match_recall_title = es.rank_eval(lisa_eval_body_match_recall,lisa_index)\n",
        "#print(json.dumps(lisa_res_match_recall, indent=4, sort_keys=True))\n",
        "lisa_create_match_recall = create_query_body_match_recall(lisa_qry_rel_data, lisa_rel, lisa_index,text_field)\n",
        "lisa_eval_body_match_recall = json.dumps(lisa_create_match_recall)\n",
        "lisa_res_match_recall_text = es.rank_eval(lisa_eval_body_match_recall,lisa_index)\n",
        "#print(json.dumps(lisa_res_match_recall, indent=4, sort_keys=True))\n",
        "lisa_create_match_recall = create_query_body_match_recall(lisa_qry_rel_data, lisa_rel, lisa_index,text_title_field)\n",
        "lisa_eval_body_match_recall = json.dumps(lisa_create_match_recall)\n",
        "lisa_res_match_recall_text_title = es.rank_eval(lisa_eval_body_match_recall,lisa_index)\n",
        "#print(json.dumps(lisa_res_match_recall, indent=4, sort_keys=True))\n",
        "\n",
        "#Medline\n",
        "med_create_match_recall = create_query_body_match_recall(med_qry_rel_data, med_rel, med_index,text_field)\n",
        "med_eval_body_match_recall = json.dumps(med_create_match_recall)\n",
        "med_res_match_recall_text = es.rank_eval(med_eval_body_match_recall, med_index)\n",
        "#print(json.dumps(med_res_match_recall, indent=4, sort_keys=True))\n",
        "\n",
        "#NPL\n",
        "npl_create_match_recall = create_query_body_match_recall(npl_qry_rel_data, npl_rel, npl_index,text_field)\n",
        "npl_eval_body_match_recall = json.dumps(npl_create_match_recall)\n",
        "npl_res_match_recall_text = es.rank_eval(npl_eval_body_match_recall,npl_index)\n",
        "#print(json.dumps(npl_res_match_recall, indent=4, sort_keys=True))\n",
        "\n",
        "#TIME\n",
        "time_create_match_recall = create_query_body_match_recall(time_qry_rel_data, time_rel, time_index,text_field)\n",
        "time_eval_body_match_recall = json.dumps(time_create_match_recall)\n",
        "time_res_match_recall_text = es.rank_eval(time_eval_body_match_recall,time_index)\n",
        "#print(json.dumps(time_res_match_recall, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IpXPiM831qRS"
      },
      "source": [
        "**BERT Sentence Embbedings On The 'title' Field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sncpWbKg1qRS",
        "outputId": "7b5e2592-dca7-41d5-c565-a19ba273288c"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "# Medline, NPL and Time corpus are excluded because they don't have any titles\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_bert_title_recall(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"recall\": {\n",
        "          \"k\" : 20,\n",
        "          \"relevant_rating_threshold\": 1,\n",
        "          #\"ignore_unlabeled\": \"false\"\n",
        "          }\n",
        "         }\n",
        "      }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\n",
        "        \"query\" : {\n",
        "  \"script_score\": {\n",
        "    \"query\": {\"match_all\": {}},\n",
        "    \"script\": {\n",
        "      \"source\": \"cosineSimilarity(params.query_vector, doc['title_vector']) + 1.0\",\n",
        "      \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "    }\n",
        "  }\n",
        "  }\n",
        "    }\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_bert_title_recall = create_query_body_bert_title_recall(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_bert_title_recall = json.dumps(adi_create_bert_title_recall)\n",
        "adi_res_bert_title_recall = es.rank_eval(adi_eval_body_bert_title_recall, adi_index)\n",
        "#print(json.dumps(adi_res_bert_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_bert_title_recall = create_query_body_bert_title_recall(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_bert_title_recall = json.dumps(cacm_create_bert_title_recall)\n",
        "cacm_res_bert_title_recall = es.rank_eval(cacm_eval_body_bert_title_recall, cacm_index)\n",
        "#print(json.dumps(cacm_res_bert_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_bert_title_recall = create_query_body_bert_title_recall(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_bert_title_recall = json.dumps(cisi_create_bert_title_recall)\n",
        "cisi_res_bert_title_recall = es.rank_eval(cisi_eval_body_bert_title_recall, cisi_index)\n",
        "#print(json.dumps(cisi_res_bert_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_bert_title_recall = create_query_body_bert_title_recall(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_bert_title_recall = json.dumps(cran_create_bert_title_recall)\n",
        "cran_res_bert_title_recall = es.rank_eval(cran_eval_body_bert_title_recall, cran_index)\n",
        "#print(json.dumps(cran_res_bert_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_bert_title_recall = create_query_body_bert_title_recall(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_bert_title_recall = json.dumps(lisa_create_bert_title_recall)\n",
        "lisa_res_bert_title_recall = es.rank_eval(lisa_eval_body_bert_title_recall, lisa_index)\n",
        "#print(json.dumps(lisa_res_bert_title_recall, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/elasticsearch/connection/base.py:200: ElasticsearchWarning: The vector functions of the form function(query, doc['field']) are deprecated, and the form function(query, 'field') should be used instead. For example, cosineSimilarity(query, doc['field']) is replaced by cosineSimilarity(query, 'field').\n",
            "  warnings.warn(message, category=ElasticsearchWarning)\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5JIrzPkG1qRT"
      },
      "source": [
        "**Re-ranking with BERT on the title field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "R_vd_1Ug1qRU",
        "outputId": "25691abf-b0a9-4fa6-a8c7-cb2c9fd8fe89"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "# Medline, NPL and Time corpus are excluded because they don't have any titles\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_rerank_title_recall(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"recall\": {\n",
        "          \"k\" : 20,\n",
        "          \"relevant_rating_threshold\": 1,\n",
        "          #\"ignore_unlabeled\": \"false\"\n",
        "          }\n",
        "         }\n",
        "      }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query =  {\n",
        "  \"query\": {\n",
        "    \"multi_match\" : {\n",
        "                  \"query\" : query_txt['question'],\n",
        "                  \"fields\" : ['title']\n",
        "            }\n",
        "  },\n",
        "  \"rescore\" : {\n",
        "      \"window_size\" : 100,\n",
        "      \"query\" : {\n",
        "         \"rescore_query\" : {\n",
        "              \"script_score\": {\n",
        "    \"query\": {\"match_all\": {}},\n",
        "    \"script\": {\n",
        "      \"source\": \"cosineSimilarity(params.query_vector, doc['title_vector']) + 1.0\",\n",
        "      \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "    }\n",
        "            }\n",
        "      }\n",
        "   }\n",
        "}\n",
        "    }\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_rerank_title_recall = create_query_body_rerank_title_recall(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_rerank_title_recall = json.dumps(adi_create_rerank_title_recall)\n",
        "adi_res_rerank_title_recall = es.rank_eval(adi_eval_body_rerank_title_recall, adi_index)\n",
        "#print(json.dumps(adi_res_rerank_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_rerank_title_recall = create_query_body_rerank_title_recall(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_rerank_title_recall = json.dumps(cacm_create_rerank_title_recall)\n",
        "cacm_res_rerank_title_recall = es.rank_eval(cacm_eval_body_rerank_title_recall, cacm_index)\n",
        "#print(json.dumps(cacm_res_rerank_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_rerank_title_recall = create_query_body_rerank_title_recall(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_rerank_title_recall = json.dumps(cisi_create_rerank_title_recall)\n",
        "cisi_res_rerank_title_recall = es.rank_eval(cisi_eval_body_rerank_title_recall, cisi_index)\n",
        "#print(json.dumps(cisi_res_rerank_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_rerank_title_recall = create_query_body_rerank_title_recall(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_rerank_title_recall = json.dumps(cran_create_rerank_title_recall)\n",
        "cran_res_rerank_title_recall = es.rank_eval(cran_eval_body_rerank_title_recall, cran_index)\n",
        "#print(json.dumps(cran_res_rerank_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_rerank_title_recall = create_query_body_rerank_title_recall(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_rerank_title_recall = json.dumps(lisa_create_rerank_title_recall)\n",
        "lisa_res_rerank_title_recall = es.rank_eval(lisa_eval_body_rerank_title_recall, lisa_index)\n",
        "#print(json.dumps(lisa_res_rerank_title_recall, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/elasticsearch/connection/base.py:200: ElasticsearchWarning: The vector functions of the form function(query, doc['field']) are deprecated, and the form function(query, 'field') should be used instead. For example, cosineSimilarity(query, doc['field']) is replaced by cosineSimilarity(query, 'field').\n",
            "  warnings.warn(message, category=ElasticsearchWarning)\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6_HuP3-p1qRX"
      },
      "source": [
        "**BERT Transformer on the 'text' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "THjLOqOG1qRX"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "med_index = 'bert-sentence-medline-corpus'\n",
        "npl_index = 'bert-sentence-npl-corpus'\n",
        "time_index = 'bert-sentence-time-corpus'\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_bert_text_recall(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"recall\": {\n",
        "          \"k\" : 20,\n",
        "          \"relevant_rating_threshold\": 1,\n",
        "          #\"ignore_unlabeled\": \"false\"\n",
        "          }\n",
        "         }\n",
        "      }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\n",
        "        \"query\": {\n",
        "            \"nested\": {\n",
        "                \"path\": \"text_vector\",\n",
        "                \"score_mode\": \"max\",\n",
        "                \"query\": {\n",
        "                    \"function_score\": {\n",
        "                        \"script_score\": {\n",
        "                            \"script\": {\n",
        "                                \"source\": \"1.0 + cosineSimilarity(params.query_vector, 'text_vector.vector')\",\n",
        "                                \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "                                }\n",
        "                                }\n",
        "                                }\n",
        "                          }\n",
        "                       }\n",
        "                  }\n",
        "                  }\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_bert_text_recall = create_query_body_bert_text_recall(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_bert_text_recall = json.dumps(adi_create_bert_text_recall)\n",
        "adi_res_bert_text_recall = es.rank_eval(adi_eval_body_bert_text_recall, adi_index)\n",
        "#print(json.dumps(adi_res, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_bert_text_recall = create_query_body_bert_text_recall(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_bert_text_recall = json.dumps(cacm_create_bert_text_recall)\n",
        "cacm_res_bert_text_recall = es.rank_eval(cacm_eval_body_bert_text_recall, cacm_index)\n",
        "#print(json.dumps(cacm_res, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_bert_text_recall = create_query_body_bert_text_recall(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_bert_text_recall = json.dumps(cisi_create_bert_text_recall)\n",
        "cisi_res_bert_text_recall = es.rank_eval(cisi_eval_body_bert_text_recall, cisi_index)\n",
        "#print(json.dumps(cisi_res, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_bert_text_recall = create_query_body_bert_text_recall(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_bert_text_recall = json.dumps(cran_create_bert_text_recall)\n",
        "cran_res_bert_text_recall = es.rank_eval(cran_eval_body_bert_text_recall, cran_index, request_timeout=160)\n",
        "#print(json.dumps(cran_res_bert_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_bert_text_recall = create_query_body_bert_text_recall(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_bert_text_recall = json.dumps(lisa_create_bert_text_recall)\n",
        "lisa_res_bert_text_recall = es.rank_eval(lisa_eval_body_bert_text_recall, lisa_index)\n",
        "#print(json.dumps(lisa_res, indent=4, sort_keys=True))\n",
        "\n",
        "# Medline\n",
        "\n",
        "med_create_bert_text_recall = create_query_body_bert_text_recall(med_qry_rel_data, med_rel, med_index)\n",
        "med_eval_body_bert_text_recall = json.dumps(med_create_bert_text_recall)\n",
        "med_res_bert_text_recall = es.rank_eval(med_eval_body_bert_text_recall, med_index)\n",
        "#print(json.dumps(med_res, indent=4, sort_keys=True))\n",
        "\n",
        "# NPL\n",
        "\n",
        "npl_create_bert_text_recall = create_query_body_bert_text_recall(npl_qry_rel_data, npl_rel, npl_index)\n",
        "npl_eval_body_bert_text_recall = json.dumps(npl_create_bert_text_recall)\n",
        "npl_res_bert_text_recall = es.rank_eval(npl_eval_body_bert_text_recall, npl_index)\n",
        "#print(json.dumps(npl_res, indent=4, sort_keys=True))\n",
        "\n",
        "# Time\n",
        "\n",
        "time_create_bert_text_recall = create_query_body_bert_text_recall(time_qry_rel_data, time_rel, time_index)\n",
        "time_eval_body_bert_text_recall = json.dumps(time_create_bert_text_recall)\n",
        "time_res_bert_text_recall = es.rank_eval(time_eval_body_bert_text_recall, time_index)\n",
        "#print(json.dumps(time_res, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "VtOMJ7Mw1qRZ"
      },
      "source": [
        "**Re-ranking with BERT on the 'text' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Y6K2BsHU1qRZ"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "med_index = 'bert-sentence-medline-corpus'\n",
        "npl_index = 'bert-sentence-npl-corpus'\n",
        "time_index = 'bert-sentence-time-corpus'\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_rerank_text_recall(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"recall\": {\n",
        "          \"k\" : 20,\n",
        "          \"relevant_rating_threshold\": 1,\n",
        "          #\"ignore_unlabeled\": \"false\"\n",
        "          }\n",
        "         }\n",
        "      }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\n",
        "  \"query\": {\n",
        "    \"multi_match\" : {\n",
        "                  \"query\" : query_txt['question'],\n",
        "                  \"fields\" : ['text']\n",
        "            }\n",
        "  },\n",
        "  \"rescore\" : {\n",
        "      \"window_size\" : 100,\n",
        "      \"query\": {\n",
        "          \"rescore_query\" : {\n",
        "            \"nested\": {\n",
        "                \"path\": \"text_vector\",\n",
        "                \"score_mode\": \"max\",\n",
        "                \"query\": {\n",
        "                    \"function_score\": {\n",
        "                        \"script_score\": {\n",
        "                            \"script\": {\n",
        "                                \"source\": \"1.0 + cosineSimilarity(params.query_vector, 'text_vector.vector')\",\n",
        "                                \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "                                }\n",
        "                                }\n",
        "                                }\n",
        "                          }\n",
        "                       }\n",
        "            }\n",
        "                  }\n",
        "}\n",
        "    }\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_rerank_text_recall = create_query_body_rerank_text_recall(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_rerank_text_recall = json.dumps(adi_create_rerank_text_recall)\n",
        "adi_res_rerank_text_recall = es.rank_eval(adi_eval_body_rerank_text_recall, adi_index)\n",
        "#print(json.dumps(adi_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_rerank_text_recall = create_query_body_rerank_text_recall(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_rerank_text_recall = json.dumps(cacm_create_rerank_text_recall)\n",
        "cacm_res_rerank_text_recall = es.rank_eval(cacm_eval_body_rerank_text_recall, cacm_index)\n",
        "#print(json.dumps(cacm_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_rerank_text_recall = create_query_body_rerank_text_recall(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_rerank_text_recall = json.dumps(cisi_create_rerank_text_recall)\n",
        "cisi_res_rerank_text_recall = es.rank_eval(cisi_eval_body_rerank_text_recall, cisi_index)\n",
        "#print(json.dumps(cisi_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_rerank_text_recall = create_query_body_rerank_text_recall(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_rerank_text_recall = json.dumps(cran_create_rerank_text_recall)\n",
        "cran_res_rerank_text_recall = es.rank_eval(cran_eval_body_rerank_text_recall, cran_index)\n",
        "#print(json.dumps(cran_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_rerank_text_recall = create_query_body_rerank_text_recall(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_rerank_text_recall = json.dumps(lisa_create_rerank_text_recall)\n",
        "lisa_res_rerank_text_recall = es.rank_eval(lisa_eval_body_rerank_text_recall, lisa_index)\n",
        "#print(json.dumps(lisa_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Medline\n",
        "\n",
        "med_create_rerank_text_recall = create_query_body_rerank_text_recall(med_qry_rel_data, med_rel, med_index)\n",
        "med_eval_body_rerank_text_recall = json.dumps(med_create_rerank_text_recall)\n",
        "med_res_rerank_text_recall = es.rank_eval(med_eval_body_rerank_text_recall, med_index)\n",
        "#print(json.dumps(med_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# NPL\n",
        "\n",
        "npl_create_rerank_text_recall = create_query_body_rerank_text_recall(npl_qry_rel_data, npl_rel, npl_index)\n",
        "npl_eval_body_rerank_text_recall = json.dumps(npl_create_rerank_text_recall)\n",
        "npl_res_rerank_text_recall = es.rank_eval(npl_eval_body_rerank_text_recall, npl_index)\n",
        "#print(json.dumps(npl_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Time\n",
        "\n",
        "time_create_rerank_text_recall = create_query_body_rerank_text_recall(time_qry_rel_data, time_rel, time_index)\n",
        "time_eval_body_rerank_text_recall = json.dumps(time_create_rerank_text_recall)\n",
        "time_res_rerank_text_recall = es.rank_eval(time_eval_body_rerank_text_recall, time_index)\n",
        "#print(json.dumps(time_res_rerank_text_recall, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Qfr9YWaU1qRa"
      },
      "source": [
        "**Complex Re-ranking Query**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "t7xAIxZf1qRb",
        "outputId": "643035c8-cddb-47cc-f908-25fc536500dd"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "med_index = 'bert-sentence-medline-corpus'\n",
        "npl_index = 'bert-sentence-npl-corpus'\n",
        "time_index = 'bert-sentence-time-corpus'\n",
        "\n",
        "query_weight = 0.7 # has to be float\n",
        "rescore_query_weight = 3.5 # has to be float\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_complex_rerank_recall(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"recall\": {\n",
        "          \"k\" : 20,\n",
        "          \"relevant_rating_threshold\": 1,\n",
        "          #\"ignore_unlabeled\": \"false\"\n",
        "          }\n",
        "         }\n",
        "      }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\n",
        "  \"query\": {\n",
        "    \"multi_match\" : {\n",
        "                  \"query\" : query_txt['question'],\n",
        "                  \"fields\" : ['text','title']\n",
        "            }\n",
        "  },\n",
        "  \"rescore\" :  [\n",
        "                {\n",
        "      \"window_size\" : 100,\n",
        "      \"query\" : {\n",
        "         \"rescore_query\" : {\n",
        "              \"script_score\": {\n",
        "    \"query\": {\"match_all\": {}},\n",
        "    \"script\": {\n",
        "      \"source\": \"cosineSimilarity(params.query_vector, doc['title_vector']) + 1.0\",\n",
        "      \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "    }\n",
        "            }\n",
        "      },\n",
        "         \"query_weight\" : query_weight,\n",
        "         \"rescore_query_weight\" : rescore_query_weight\n",
        "   }\n",
        "},\n",
        "                {\n",
        "      \"window_size\" : 100,\n",
        "      \"query\": {\n",
        "          \"rescore_query\" : {\n",
        "            \"nested\": {\n",
        "                \"path\": \"text_vector\",\n",
        "                \"score_mode\": \"max\",\n",
        "                \"query\": {\n",
        "                    \"function_score\": {\n",
        "                        \"script_score\": {\n",
        "                            \"script\": {\n",
        "                                \"source\": \"1.0 + cosineSimilarity(params.query_vector, 'text_vector.vector')\",\n",
        "                                \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "                                }\n",
        "                                }\n",
        "                                }\n",
        "                          }\n",
        "                       }\n",
        "            }\n",
        "                  }\n",
        "}\n",
        "  ]}\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_complex_rerank_recall = create_query_body_complex_rerank_recall(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_complex_rerank_recall = json.dumps(adi_create_complex_rerank_recall)\n",
        "adi_res_complex_rerank_recall = es.rank_eval(adi_eval_body_complex_rerank_recall, adi_index)\n",
        "#print(json.dumps(adi_res_complex_rerank_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_complex_rerank_recall = create_query_body_complex_rerank_recall(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_complex_rerank_recall = json.dumps(cacm_create_complex_rerank_recall)\n",
        "cacm_res_complex_rerank_recall = es.rank_eval(cacm_eval_body_complex_rerank_recall, cacm_index)\n",
        "#print(json.dumps(cacm_res_complex_rerank_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_complex_rerank_recall = create_query_body_complex_rerank_recall(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_complex_rerank_recall = json.dumps(cisi_create_complex_rerank_recall)\n",
        "cisi_res_complex_rerank_recall = es.rank_eval(cisi_eval_body_complex_rerank_recall, cisi_index)\n",
        "#print(json.dumps(cisi_res_complex_rerank_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_complex_rerank_recall = create_query_body_complex_rerank_recall(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_complex_rerank_recall = json.dumps(cran_create_complex_rerank_recall)\n",
        "cran_res_complex_rerank_recall = es.rank_eval(cran_eval_body_complex_rerank_recall, cran_index)\n",
        "#print(json.dumps(cran_res_complex_rerank_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_complex_rerank_recall = create_query_body_complex_rerank_recall(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_complex_rerank_recall = json.dumps(lisa_create_complex_rerank_recall)\n",
        "lisa_res_complex_rerank_recall = es.rank_eval(lisa_eval_body_complex_rerank_recall, lisa_index)\n",
        "#print(json.dumps(lisa_res_complex_rerank_recall, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/elasticsearch/connection/base.py:200: ElasticsearchWarning: The vector functions of the form function(query, doc['field']) are deprecated, and the form function(query, 'field') should be used instead. For example, cosineSimilarity(query, doc['field']) is replaced by cosineSimilarity(query, 'field').\n",
            "  warnings.warn(message, category=ElasticsearchWarning)\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ewr2jzrL1qRc"
      },
      "source": [
        "### Precision\n",
        "\n",
        "In this section, we only evaluate the Precision scores."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "QjI4qvB61qRd"
      },
      "source": [
        "**Multi Match Query**\n",
        "\n",
        "Here, we evaluate the data with the [\"multi_match\"](https://pragmalingu.de/docs/experiments/experiment1#standard-elasticsearch) option of Elasticsearch:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rmjO-NDn1qRd"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/query-dsl-simple-query-string-query.html\n",
        "\n",
        "import json\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'adi-corpus'\n",
        "cacm_index = 'cacm-corpus'\n",
        "cisi_index = 'cisi-corpus'\n",
        "cran_index = 'cranfield-corpus'\n",
        "lisa_index = 'lisa-corpus'\n",
        "med_index = 'medline-corpus'\n",
        "npl_index = 'npl-corpus'\n",
        "time_index = 'time-corpus'\n",
        "\n",
        "text_field = ['text']\n",
        "title_field = ['title']\n",
        "text_title_field = ['title','text']\n",
        "\n",
        "\n",
        "#function to get normal match evaluation body\n",
        "def create_query_body_match_precision(query_rel_dict, rel_dict, index_name, field):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\": '',\n",
        "      \"metric\": {\n",
        "          \"precision\": {\n",
        "              \"relevant_rating_threshold\": 1,\n",
        "              \"k\": 20\n",
        "              }\n",
        "      }\n",
        "  }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\"query\": { \"multi_match\": { \"query\": '' , \"fields\" : field}}}\n",
        "    current_query[\"query\"][\"multi_match\"][\"query\"] = query_txt['question']\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\"_index\": index_name, \"_id\": str(el), \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "\n",
        "#ADI\n",
        "adi_create_match_precision= create_query_body_match_precision(adi_qry_rel_data, adi_rel, adi_index,title_field)\n",
        "adi_eval_body_match_precision = json.dumps(adi_create_match_precision)\n",
        "adi_res_match_precision_title = es.rank_eval(adi_eval_body_match_precision, adi_index)\n",
        "#print(json.dumps(adi_res_match_precision, indent=4, sort_keys=True))\n",
        "adi_create_match_precision= create_query_body_match_precision(adi_qry_rel_data, adi_rel, adi_index,text_field)\n",
        "adi_eval_body_match_precision = json.dumps(adi_create_match_precision)\n",
        "adi_res_match_precision_text = es.rank_eval(adi_eval_body_match_precision, adi_index)\n",
        "#print(json.dumps(adi_res_match_precision, indent=4, sort_keys=True))\n",
        "adi_create_match_precision= create_query_body_match_precision(adi_qry_rel_data, adi_rel, adi_index,text_title_field)\n",
        "adi_eval_body_match_precision = json.dumps(adi_create_match_precision)\n",
        "adi_res_match_precision_text_title = es.rank_eval(adi_eval_body_match_precision, adi_index)\n",
        "#print(json.dumps(adi_res_match_precision, indent=4, sort_keys=True))\n",
        "\n",
        "#CACM\n",
        "cacm_create_match_precision = create_query_body_match_precision(cacm_qry_rel_data, cacm_rel, cacm_index,title_field)\n",
        "cacm_eval_body_match_precision = json.dumps(cacm_create_match_precision)\n",
        "cacm_res_match_precision_title = es.rank_eval(cacm_eval_body_match_precision,cacm_index)\n",
        "#print(json.dumps(cacm_res_match_precision, indent=4, sort_keys=True))\n",
        "cacm_create_match_precision = create_query_body_match_precision(cacm_qry_rel_data, cacm_rel, cacm_index,text_field)\n",
        "cacm_eval_body_match_precision = json.dumps(cacm_create_match_precision)\n",
        "cacm_res_match_precision_text = es.rank_eval(cacm_eval_body_match_precision,cacm_index)\n",
        "#print(json.dumps(cacm_res_match_precision, indent=4, sort_keys=True))\n",
        "cacm_create_match_precision = create_query_body_match_precision(cacm_qry_rel_data, cacm_rel, cacm_index,text_title_field)\n",
        "cacm_eval_body_match_precision = json.dumps(cacm_create_match_precision)\n",
        "cacm_res_match_precision_text_title = es.rank_eval(cacm_eval_body_match_precision,cacm_index)\n",
        "#print(json.dumps(cacm_res_match_precision, indent=4, sort_keys=True))\n",
        "\n",
        "#CISI\n",
        "cisi_create_match_precision = create_query_body_match_precision(cisi_qry_rel_data, cisi_rel, cisi_index,title_field)\n",
        "cisi_eval_body_match_precision = json.dumps(cisi_create_match_precision)\n",
        "cisi_res_match_precision_title = es.rank_eval(cisi_eval_body_match_precision, cisi_index)\n",
        "#print(json.dumps(cisi_res_match_precision, indent=4, sort_keys=True))\n",
        "cisi_create_match_precision = create_query_body_match_precision(cisi_qry_rel_data, cisi_rel, cisi_index,text_field)\n",
        "cisi_eval_body_match_precision = json.dumps(cisi_create_match_precision)\n",
        "cisi_res_match_precision_text = es.rank_eval(cisi_eval_body_match_precision, cisi_index)\n",
        "#print(json.dumps(cisi_res_match_precision, indent=4, sort_keys=True))\n",
        "cisi_create_match_precision = create_query_body_match_precision(cisi_qry_rel_data, cisi_rel, cisi_index,text_title_field)\n",
        "cisi_eval_body_match_precision = json.dumps(cisi_create_match_precision)\n",
        "cisi_res_match_precision_text_title = es.rank_eval(cisi_eval_body_match_precision, cisi_index)\n",
        "#print(json.dumps(cisi_res_match_precision, indent=4, sort_keys=True))\n",
        "\n",
        "#Cranfield\n",
        "cran_create_match_precision = create_query_body_match_precision(cran_qry_rel_data, cran_rel, cran_index,title_field)\n",
        "cran_eval_body_match_precision = json.dumps(cran_create_match_precision)\n",
        "cran_res_match_precision_title = es.rank_eval(cran_eval_body_match_precision, cran_index)\n",
        "#print(json.dumps(cran_res_match_precision, indent=4, sort_keys=True))\n",
        "cran_create_match_precision = create_query_body_match_precision(cran_qry_rel_data, cran_rel, cran_index,text_field)\n",
        "cran_eval_body_match_precision = json.dumps(cran_create_match_precision)\n",
        "cran_res_match_precision_text = es.rank_eval(cran_eval_body_match_precision, cran_index)\n",
        "#print(json.dumps(cran_res_match_precision, indent=4, sort_keys=True))\n",
        "cran_create_match_precision = create_query_body_match_precision(cran_qry_rel_data, cran_rel, cran_index,text_title_field)\n",
        "cran_eval_body_match_precision = json.dumps(cran_create_match_precision)\n",
        "cran_res_match_precision_text_title = es.rank_eval(cran_eval_body_match_precision, cran_index)\n",
        "#print(json.dumps(cran_res_match_precision, indent=4, sort_keys=True))\n",
        "\n",
        "#LISA\n",
        "lisa_create_match_precision = create_query_body_match_precision(lisa_qry_rel_data, lisa_rel, lisa_index,title_field)\n",
        "lisa_eval_body_match_precision = json.dumps(lisa_create_match_precision)\n",
        "lisa_res_match_precision_title = es.rank_eval(lisa_eval_body_match_precision,lisa_index)\n",
        "#print(json.dumps(lisa_res_match_precision, indent=4, sort_keys=True))\n",
        "lisa_create_match_precision = create_query_body_match_precision(lisa_qry_rel_data, lisa_rel, lisa_index,text_field)\n",
        "lisa_eval_body_match_precision = json.dumps(lisa_create_match_precision)\n",
        "lisa_res_match_precision_text = es.rank_eval(lisa_eval_body_match_precision,lisa_index)\n",
        "#print(json.dumps(lisa_res_match_precision, indent=4, sort_keys=True))\n",
        "lisa_create_match_precision = create_query_body_match_precision(lisa_qry_rel_data, lisa_rel, lisa_index,text_title_field)\n",
        "lisa_eval_body_match_precision = json.dumps(lisa_create_match_precision)\n",
        "lisa_res_match_precision_text_title = es.rank_eval(lisa_eval_body_match_precision,lisa_index)\n",
        "#print(json.dumps(lisa_res_match_precision, indent=4, sort_keys=True))\n",
        "\n",
        "#Medline\n",
        "med_create_match_precision = create_query_body_match_precision(med_qry_rel_data, med_rel, med_index,text_field)\n",
        "med_eval_body_match_precision = json.dumps(med_create_match_precision)\n",
        "med_res_match_precision_text = es.rank_eval(med_eval_body_match_precision, med_index)\n",
        "#print(json.dumps(med_res_match_precision, indent=4, sort_keys=True))\n",
        "\n",
        "#NPL\n",
        "npl_create_match_precision = create_query_body_match_precision(npl_qry_rel_data, npl_rel, npl_index,text_field)\n",
        "npl_eval_body_match_precision = json.dumps(npl_create_match_precision)\n",
        "npl_res_match_precision_text = es.rank_eval(npl_eval_body_match_precision,npl_index)\n",
        "#print(json.dumps(npl_res_match_precision, indent=4, sort_keys=True))\n",
        "\n",
        "#TIME\n",
        "time_create_match_precision = create_query_body_match_precision(time_qry_rel_data, time_rel, time_index,text_field)\n",
        "time_eval_body_match_precision = json.dumps(time_create_match_precision)\n",
        "time_res_match_precision_text = es.rank_eval(time_eval_body_match_precision,time_index)\n",
        "#print(json.dumps(time_res_match_precision, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dSWiUSxK1qRf"
      },
      "source": [
        "**BERT Sentence Embeddings on the 'title' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "sKxUX2r81qRg",
        "outputId": "fc16b252-d5be-4a13-92aa-7aaddf84a962"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "# Medline, NPL and Time corpus are excluded because they don't have any titles\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_bert_title_precision(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"precision\": {\n",
        "            \"k\" : 20,\n",
        "            \"relevant_rating_threshold\": 1,\n",
        "            #\"ignore_unlabeled\": \"false\"\n",
        "            }\n",
        "      }\n",
        "  }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\n",
        "        \"query\" : {\n",
        "  \"script_score\": {\n",
        "    \"query\": {\"match_all\": {}},\n",
        "    \"script\": {\n",
        "      \"source\": \"cosineSimilarity(params.query_vector, doc['title_vector']) + 1.0\",\n",
        "      \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "    }\n",
        "  }\n",
        "  }\n",
        "    }\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_bert_title_precision = create_query_body_bert_title_recall(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_bert_title_precision = json.dumps(adi_create_bert_title_precision)\n",
        "adi_res_bert_title_precision = es.rank_eval(adi_eval_body_bert_title_precision, adi_index)\n",
        "#print(json.dumps(adi_res_bert_title_precision, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_bert_title_precision = create_query_body_bert_title_precision(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_bert_title_precision = json.dumps(cacm_create_bert_title_precision)\n",
        "cacm_res_bert_title_precision = es.rank_eval(cacm_eval_body_bert_title_precision, cacm_index)\n",
        "#print(json.dumps(cacm_res_bert_title_precision, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_bert_title_precision = create_query_body_bert_title_precision(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_bert_title_precision = json.dumps(cisi_create_bert_title_precision)\n",
        "cisi_res_bert_title_precision = es.rank_eval(cisi_eval_body_bert_title_precision, cisi_index)\n",
        "#print(json.dumps(cisi_res_bert_title_precision, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_bert_title_precision = create_query_body_bert_title_precision(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_bert_title_precision = json.dumps(cran_create_bert_title_precision)\n",
        "cran_res_bert_title_precision = es.rank_eval(cran_eval_body_bert_title_precision, cran_index)\n",
        "#print(json.dumps(cran_res_bert_title_precision, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_bert_title_precision = create_query_body_bert_title_precision(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_bert_title_precision = json.dumps(lisa_create_bert_title_precision)\n",
        "lisa_res_bert_title_precision = es.rank_eval(lisa_eval_body_bert_title_precision, lisa_index)\n",
        "#print(json.dumps(lisa_res_bert_title_precision, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/elasticsearch/connection/base.py:200: ElasticsearchWarning: The vector functions of the form function(query, doc['field']) are deprecated, and the form function(query, 'field') should be used instead. For example, cosineSimilarity(query, doc['field']) is replaced by cosineSimilarity(query, 'field').\n",
            "  warnings.warn(message, category=ElasticsearchWarning)\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "WNAUq53P1qRh"
      },
      "source": [
        "**Re-ranking with BERT on the 'title' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "BRwxhcmW1qRh",
        "outputId": "84d11652-f820-497f-bb42-9d05cedefcd6"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "# Medline, NPL and Time corpus are excluded because they don't have any titles\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_rerank_title_precision(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"precision\": {\n",
        "          \"k\" : 20,\n",
        "          \"relevant_rating_threshold\": 1,\n",
        "          #\"ignore_unlabeled\": \"false\"\n",
        "          }\n",
        "         }\n",
        "      }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query =  {\n",
        "  \"query\": {\n",
        "    \"multi_match\" : {\n",
        "                  \"query\" : query_txt['question'],\n",
        "                  \"fields\" : ['title']\n",
        "            }\n",
        "  },\n",
        "  \"rescore\" : {\n",
        "      \"window_size\" : 100,\n",
        "      \"query\" : {\n",
        "         \"rescore_query\" : {\n",
        "              \"script_score\": {\n",
        "    \"query\": {\"match_all\": {}},\n",
        "    \"script\": {\n",
        "      \"source\": \"cosineSimilarity(params.query_vector, doc['title_vector']) + 1.0\",\n",
        "      \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "    }\n",
        "            }\n",
        "      }\n",
        "   }\n",
        "}\n",
        "    }\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_rerank_title_precision = create_query_body_rerank_title_precision(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_rerank_title_precision = json.dumps(adi_create_rerank_title_precision)\n",
        "adi_res_rerank_title_precision = es.rank_eval(adi_eval_body_rerank_title_precision, adi_index)\n",
        "#print(json.dumps(adi_res_rerank_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_rerank_title_precision = create_query_body_rerank_title_precision(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_rerank_title_precision = json.dumps(cacm_create_rerank_title_precision)\n",
        "cacm_res_rerank_title_precision = es.rank_eval(cacm_eval_body_rerank_title_precision, cacm_index)\n",
        "#print(json.dumps(cacm_res_rerank_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_rerank_title_precision = create_query_body_rerank_title_precision(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_rerank_title_precision = json.dumps(cisi_create_rerank_title_precision)\n",
        "cisi_res_rerank_title_precision = es.rank_eval(cisi_eval_body_rerank_title_precision, cisi_index)\n",
        "#print(json.dumps(cisi_res_rerank_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_rerank_title_precision = create_query_body_rerank_title_precision(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_rerank_title_precision = json.dumps(cran_create_rerank_title_precision)\n",
        "cran_res_rerank_title_precision = es.rank_eval(cran_eval_body_rerank_title_precision, cran_index)\n",
        "#print(json.dumps(cran_res_rerank_title_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_rerank_title_precision = create_query_body_rerank_title_precision(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_rerank_title_precision = json.dumps(lisa_create_rerank_title_precision)\n",
        "lisa_res_rerank_title_precision = es.rank_eval(lisa_eval_body_rerank_title_precision, lisa_index)\n",
        "#print(json.dumps(lisa_res_rerank_title_recall, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/elasticsearch/connection/base.py:200: ElasticsearchWarning: The vector functions of the form function(query, doc['field']) are deprecated, and the form function(query, 'field') should be used instead. For example, cosineSimilarity(query, doc['field']) is replaced by cosineSimilarity(query, 'field').\n",
            "  warnings.warn(message, category=ElasticsearchWarning)\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "w597XyY61qRi"
      },
      "source": [
        "**BERT Transformer on the 'text' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lzEF1NqC1qRi"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "med_index = 'bert-sentence-medline-corpus'\n",
        "npl_index = 'bert-sentence-npl-corpus'\n",
        "time_index = 'bert-sentence-time-corpus'\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_bert_text_precision(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"precision\": {\n",
        "          \"k\" : 20,\n",
        "          \"relevant_rating_threshold\": 1,\n",
        "          #\"ignore_unlabeled\": \"false\"\n",
        "          }\n",
        "         }\n",
        "      }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\n",
        "        \"query\": {\n",
        "            \"nested\": {\n",
        "                \"path\": \"text_vector\",\n",
        "                \"score_mode\": \"max\",\n",
        "                \"query\": {\n",
        "                    \"function_score\": {\n",
        "                        \"script_score\": {\n",
        "                            \"script\": {\n",
        "                                \"source\": \"1.0 + cosineSimilarity(params.query_vector, 'text_vector.vector')\",\n",
        "                                \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "                                }\n",
        "                                }\n",
        "                                }\n",
        "                          }\n",
        "                       }\n",
        "                  }\n",
        "                  }\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_bert_text_precision = create_query_body_bert_text_precision(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_bert_text_precision = json.dumps(adi_create_bert_text_precision)\n",
        "adi_res_bert_text_precision = es.rank_eval(adi_eval_body_bert_text_precision, adi_index)\n",
        "#print(json.dumps(adi_res, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_bert_text_precision = create_query_body_bert_text_precision(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_bert_text_precision = json.dumps(cacm_create_bert_text_precision)\n",
        "cacm_res_bert_text_precision = es.rank_eval(cacm_eval_body_bert_text_precision, cacm_index)\n",
        "#print(json.dumps(cacm_res, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_bert_text_precision = create_query_body_bert_text_precision(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_bert_text_precision = json.dumps(cisi_create_bert_text_precision)\n",
        "cisi_res_bert_text_precision = es.rank_eval(cisi_eval_body_bert_text_precision, cisi_index)\n",
        "#print(json.dumps(cisi_res, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_bert_text_precision = create_query_body_bert_text_precision(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_bert_text_precision = json.dumps(cran_create_bert_text_precision)\n",
        "cran_res_bert_text_precision = es.rank_eval(cran_eval_body_bert_text_precision, cran_index)\n",
        "#print(json.dumps(cran_res, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_bert_text_precision = create_query_body_bert_text_precision(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_bert_text_precision = json.dumps(lisa_create_bert_text_precision)\n",
        "lisa_res_bert_text_precision = es.rank_eval(lisa_eval_body_bert_text_precision, lisa_index)\n",
        "#print(json.dumps(lisa_res, indent=4, sort_keys=True))\n",
        "\n",
        "# Medline\n",
        "\n",
        "med_create_bert_text_precision = create_query_body_bert_text_precision(med_qry_rel_data, med_rel, med_index)\n",
        "med_eval_body_bert_text_precision = json.dumps(med_create_bert_text_precision)\n",
        "med_res_bert_text_precision = es.rank_eval(med_eval_body_bert_text_precision, med_index)\n",
        "#print(json.dumps(med_res, indent=4, sort_keys=True))\n",
        "\n",
        "# NPL\n",
        "\n",
        "npl_create_bert_text_precision = create_query_body_bert_text_precision(npl_qry_rel_data, npl_rel, npl_index)\n",
        "npl_eval_body_bert_text_precision = json.dumps(npl_create_bert_text_precision)\n",
        "npl_res_bert_text_precision = es.rank_eval(npl_eval_body_bert_text_precision, npl_index)\n",
        "#print(json.dumps(npl_res, indent=4, sort_keys=True))\n",
        "\n",
        "# Time\n",
        "\n",
        "time_create_bert_text_precision = create_query_body_bert_text_precision(time_qry_rel_data, time_rel, time_index)\n",
        "time_eval_body_bert_text_precision = json.dumps(time_create_bert_text_precision)\n",
        "time_res_bert_text_precision = es.rank_eval(time_eval_body_bert_text_precision, time_index)\n",
        "#print(json.dumps(time_res, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "m6EzDo3S1qRk"
      },
      "source": [
        "**Re-ranking with BERT on the 'text' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eturzIWD1qRk"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "med_index = 'bert-sentence-medline-corpus'\n",
        "npl_index = 'bert-sentence-npl-corpus'\n",
        "time_index = 'bert-sentence-time-corpus'\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_rerank_text_precision(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"precision\": {\n",
        "          \"k\" : 20,\n",
        "          \"relevant_rating_threshold\": 1,\n",
        "          #\"ignore_unlabeled\": \"false\"\n",
        "          }\n",
        "         }\n",
        "      }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\n",
        "  \"query\": {\n",
        "    \"multi_match\" : {\n",
        "                  \"query\" : query_txt['question'],\n",
        "                  \"fields\" : ['text']\n",
        "            }\n",
        "  },\n",
        "  \"rescore\" : {\n",
        "      \"window_size\" : 100,\n",
        "      \"query\": {\n",
        "          \"rescore_query\" : {\n",
        "            \"nested\": {\n",
        "                \"path\": \"text_vector\",\n",
        "                \"score_mode\": \"max\",\n",
        "                \"query\": {\n",
        "                    \"function_score\": {\n",
        "                        \"script_score\": {\n",
        "                            \"script\": {\n",
        "                                \"source\": \"1.0 + cosineSimilarity(params.query_vector, 'text_vector.vector')\",\n",
        "                                \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "                                }\n",
        "                                }\n",
        "                                }\n",
        "                          }\n",
        "                       }\n",
        "            }\n",
        "                  }\n",
        "}\n",
        "    }\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_rerank_text_precision = create_query_body_rerank_text_precision(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_rerank_text_precision = json.dumps(adi_create_rerank_text_precision)\n",
        "adi_res_rerank_text_precision = es.rank_eval(adi_eval_body_rerank_text_precision, adi_index)\n",
        "#print(json.dumps(adi_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_rerank_text_precision = create_query_body_rerank_text_precision(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_rerank_text_precision = json.dumps(cacm_create_rerank_text_precision)\n",
        "cacm_res_rerank_text_precision = es.rank_eval(cacm_eval_body_rerank_text_precision, cacm_index)\n",
        "#print(json.dumps(cacm_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_rerank_text_precision = create_query_body_rerank_text_precision(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_rerank_text_precision = json.dumps(cisi_create_rerank_text_precision)\n",
        "cisi_res_rerank_text_precision = es.rank_eval(cisi_eval_body_rerank_text_precision, cisi_index)\n",
        "#print(json.dumps(cisi_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_rerank_text_precision = create_query_body_rerank_text_precision(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_rerank_text_precision = json.dumps(cran_create_rerank_text_precision)\n",
        "cran_res_rerank_text_precision = es.rank_eval(cran_eval_body_rerank_text_precision, cran_index)\n",
        "#print(json.dumps(cran_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_rerank_text_precision = create_query_body_rerank_text_precision(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_rerank_text_precision = json.dumps(lisa_create_rerank_text_precision)\n",
        "lisa_res_rerank_text_precision = es.rank_eval(lisa_eval_body_rerank_text_precision, lisa_index)\n",
        "#print(json.dumps(lisa_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Medline\n",
        "\n",
        "med_create_rerank_text_precision = create_query_body_rerank_text_precision(med_qry_rel_data, med_rel, med_index)\n",
        "med_eval_body_rerank_text_precision = json.dumps(med_create_rerank_text_precision)\n",
        "med_res_rerank_text_precision = es.rank_eval(med_eval_body_rerank_text_precision, med_index)\n",
        "#print(json.dumps(med_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# NPL\n",
        "\n",
        "npl_create_rerank_text_precision = create_query_body_rerank_text_precision(npl_qry_rel_data, npl_rel, npl_index)\n",
        "npl_eval_body_rerank_text_precision = json.dumps(npl_create_rerank_text_precision)\n",
        "npl_res_rerank_text_precision = es.rank_eval(npl_eval_body_rerank_text_precision, npl_index)\n",
        "#print(json.dumps(npl_res_rerank_text_recall, indent=4, sort_keys=True))\n",
        "\n",
        "# Time\n",
        "\n",
        "time_create_rerank_text_precision = create_query_body_rerank_text_precision(time_qry_rel_data, time_rel, time_index)\n",
        "time_eval_body_rerank_text_precision = json.dumps(time_create_rerank_text_precision)\n",
        "time_res_rerank_text_precision = es.rank_eval(time_eval_body_rerank_text_precision, time_index)\n",
        "#print(json.dumps(time_res_rerank_text_recall, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IoRCTe3S1qRl"
      },
      "source": [
        "**Complex Re-ranking Query**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "rRjOJOYP1qRl",
        "outputId": "48597658-0f23-4e74-bef8-164319f354cd"
      },
      "source": [
        "#use rank eval api, see https://elasticsearch-py.readthedocs.io/en/master/api.html?highlight=_rank_eval#elasticsearch.Elasticsearch.rank_eval\n",
        "#and https://www.elastic.co/guide/en/elasticsearch/reference/current/search-rank-eval.html#search-rank-eval\n",
        "\n",
        "from collections import defaultdict\n",
        "\n",
        "adi_index = 'bert-sentence-adi-corpus'\n",
        "cacm_index = 'bert-sentence-cacm-corpus'\n",
        "cisi_index = 'bert-sentence-cisi-corpus'\n",
        "cran_index = 'bert-sentence-cranfield-corpus'\n",
        "lisa_index = 'bert-sentence-lisa-corpus'\n",
        "med_index = 'bert-sentence-medline-corpus'\n",
        "npl_index = 'bert-sentence-npl-corpus'\n",
        "time_index = 'bert-sentence-time-corpus'\n",
        "\n",
        "query_weight = 0.7 # has to be float\n",
        "rescore_query_weight = 3.5 # has to be float\n",
        "\n",
        "# function to get normal match evaluation body\n",
        "\n",
        "def create_query_body_complex_rerank_precision(query_rel_dict, rel_dict, index_name):\n",
        "  \"\"\"\n",
        "  The function creates a request for every query in query_dict and rates the relevant documents with rel_dict to 1.\n",
        "  The index name has to be the same as from the documents your looking at.\n",
        "  An evaluation body for the elasticsearch ranking API is returned.\n",
        "  \"\"\"\n",
        "  eval_body = {\n",
        "      \"requests\":'',\n",
        "      \"metric\": {\n",
        "          \"precision\": {\n",
        "          \"k\" : 20,\n",
        "          \"relevant_rating_threshold\": 1,\n",
        "          #\"ignore_unlabeled\": \"false\"\n",
        "          }\n",
        "         }\n",
        "      }\n",
        "  requests = []\n",
        "  current_request = defaultdict(lambda: defaultdict())\n",
        "  current_rel = {\"_index\": index_name, \"_id\": '', \"rating\": int}\n",
        "  for query_ID, query_txt in query_rel_dict.items():\n",
        "    current_query = {\n",
        "  \"query\": {\n",
        "    \"multi_match\" : {\n",
        "                  \"query\" : query_txt['question'],\n",
        "                  \"fields\" : ['text','title']\n",
        "            }\n",
        "  },\n",
        "  \"rescore\" :  [\n",
        "                {\n",
        "      \"window_size\" : 100,\n",
        "      \"query\" : {\n",
        "         \"rescore_query\" : {\n",
        "              \"script_score\": {\n",
        "    \"query\": {\"match_all\": {}},\n",
        "    \"script\": {\n",
        "      \"source\": \"cosineSimilarity(params.query_vector, doc['title_vector']) + 1.0\",\n",
        "      \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "    }\n",
        "            }\n",
        "      },\n",
        "         \"query_weight\" : query_weight,\n",
        "         \"rescore_query_weight\" : rescore_query_weight\n",
        "   }\n",
        "},\n",
        "                {\n",
        "      \"window_size\" : 100,\n",
        "      \"query\": {\n",
        "          \"rescore_query\" : {\n",
        "            \"nested\": {\n",
        "                \"path\": \"text_vector\",\n",
        "                \"score_mode\": \"max\",\n",
        "                \"query\": {\n",
        "                    \"function_score\": {\n",
        "                        \"script_score\": {\n",
        "                            \"script\": {\n",
        "                                \"source\": \"1.0 + cosineSimilarity(params.query_vector, 'text_vector.vector')\",\n",
        "                                \"params\": {\"query_vector\": list(model.encode(query_txt['question']).astype(float))}\n",
        "                                }\n",
        "                                }\n",
        "                                }\n",
        "                          }\n",
        "                       }\n",
        "            }\n",
        "                  }\n",
        "}\n",
        "  ]}\n",
        "    current_request[\"id\"] = 'Query_'+str(query_ID)\n",
        "    current_request[\"request\"] = current_query.copy()\n",
        "    current_request[\"ratings\"] = [{\n",
        "          \"_index\": index_name,\n",
        "          \"_id\": str(el) if type(el) != tuple else str(el[0]),\n",
        "          \"rating\": 1} for el in rel_dict[query_ID]]\n",
        "    requests.append(current_request.copy())\n",
        "  eval_body[\"requests\"] = requests\n",
        "  return eval_body\n",
        "\n",
        "# ADI\n",
        "\n",
        "adi_create_complex_rerank_precision = create_query_body_complex_rerank_precision(adi_qry_rel_data, adi_rel, adi_index)\n",
        "adi_eval_body_complex_rerank_precision = json.dumps(adi_create_complex_rerank_precision)\n",
        "adi_res_complex_rerank_precision = es.rank_eval(adi_eval_body_complex_rerank_precision, adi_index)\n",
        "#print(json.dumps(adi_res_complex_rerank_precision, indent=4, sort_keys=True))\n",
        "\n",
        "# CACM\n",
        "\n",
        "cacm_create_complex_rerank_precision = create_query_body_complex_rerank_precision(cacm_qry_rel_data, cacm_rel, cacm_index)\n",
        "cacm_eval_body_complex_rerank_precision = json.dumps(cacm_create_complex_rerank_precision)\n",
        "cacm_res_complex_rerank_precision = es.rank_eval(cacm_eval_body_complex_rerank_precision, cacm_index)\n",
        "#print(json.dumps(cacm_res_complex_rerank_precision, indent=4, sort_keys=True))\n",
        "\n",
        "# CISI\n",
        "\n",
        "cisi_create_complex_rerank_precision = create_query_body_complex_rerank_precision(cisi_qry_rel_data, cisi_rel, cisi_index)\n",
        "cisi_eval_body_complex_rerank_precision = json.dumps(cisi_create_complex_rerank_precision)\n",
        "cisi_res_complex_rerank_precision = es.rank_eval(cisi_eval_body_complex_rerank_precision, cisi_index)\n",
        "#print(json.dumps(cisi_res_complex_rerank_precision, indent=4, sort_keys=True))\n",
        "\n",
        "# Cranfield\n",
        "\n",
        "cran_create_complex_rerank_precision = create_query_body_complex_rerank_precision(cran_qry_rel_data, cran_rel, cran_index)\n",
        "cran_eval_body_complex_rerank_precision = json.dumps(cran_create_complex_rerank_precision)\n",
        "cran_res_complex_rerank_precision = es.rank_eval(cran_eval_body_complex_rerank_precision, cran_index)\n",
        "#print(json.dumps(cran_res_complex_rerank_precision, indent=4, sort_keys=True))\n",
        "\n",
        "# LISA\n",
        "\n",
        "lisa_create_complex_rerank_precision = create_query_body_complex_rerank_precision(lisa_qry_rel_data, lisa_rel, lisa_index)\n",
        "lisa_eval_body_complex_rerank_precision = json.dumps(lisa_create_complex_rerank_precision)\n",
        "lisa_res_complex_rerank_precision = es.rank_eval(lisa_eval_body_complex_rerank_precision, lisa_index)\n",
        "#print(json.dumps(lisa_res_complex_rerank_precision, indent=4, sort_keys=True))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.7/dist-packages/elasticsearch/connection/base.py:200: ElasticsearchWarning: The vector functions of the form function(query, doc['field']) are deprecated, and the form function(query, 'field') should be used instead. For example, cosineSimilarity(query, doc['field']) is replaced by cosineSimilarity(query, 'field').\n",
            "  warnings.warn(message, category=ElasticsearchWarning)\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MNV3pjat1qRm"
      },
      "source": [
        "## Visualisation\n",
        "\n",
        "The last step, is to visualize the data so we can analyze the differences:"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SaAyoRau1qRm"
      },
      "source": [
        "### Recall"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6k89jXOc1qRm"
      },
      "source": [
        "**Recall on the 'title' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 647
        },
        "id": "N7rn1T-q1qRn",
        "outputId": "253d658a-3f7e-4b2d-9461-fc106fcee35a"
      },
      "source": [
        "import matplotlib\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "match_metrics_recall_title = []\n",
        "match_metrics_recall_title.append(round(adi_res_match_recall_title['metric_score'], 3))\n",
        "match_metrics_recall_title.append(round(cacm_res_match_recall_title['metric_score'], 3))\n",
        "match_metrics_recall_title.append(round(cisi_res_match_recall_title['metric_score'], 3))\n",
        "match_metrics_recall_title.append(round(cran_res_match_recall_title['metric_score'], 3))\n",
        "match_metrics_recall_title.append(round(lisa_res_match_recall_title['metric_score'], 3))\n",
        "\n",
        "bert_title_metrics_recall = []\n",
        "bert_title_metrics_recall.append(round(adi_res_bert_title_recall['metric_score'], 3))\n",
        "bert_title_metrics_recall.append(round(cacm_res_bert_title_recall['metric_score'], 3))\n",
        "bert_title_metrics_recall.append(round(cisi_res_bert_title_recall['metric_score'], 3))\n",
        "bert_title_metrics_recall.append(round(cran_res_bert_title_recall['metric_score'], 3))\n",
        "bert_title_metrics_recall.append(round(lisa_res_bert_title_recall['metric_score'], 3))\n",
        "\n",
        "rerank_bert_title_metrics_recall = []\n",
        "rerank_bert_title_metrics_recall.append(round(adi_res_rerank_title_recall['metric_score'], 3))\n",
        "rerank_bert_title_metrics_recall.append(round(cacm_res_rerank_title_recall['metric_score'], 3))\n",
        "rerank_bert_title_metrics_recall.append(round(cisi_res_rerank_title_recall['metric_score'], 3))\n",
        "rerank_bert_title_metrics_recall.append(round(cran_res_rerank_title_recall['metric_score'], 3))\n",
        "rerank_bert_title_metrics_recall.append(round(lisa_res_rerank_title_recall['metric_score'], 3))\n",
        "\n",
        "labels = ['ADI','CACM', 'CISI','Cranfield', 'LISA']\n",
        "\n",
        "x = np.arange(len(labels))*2.5  # the label locations\n",
        "\n",
        "width = 0.5  # the width of the bars\n",
        "\n",
        "# orange, green, turquoise, black, red, yellow, white\n",
        "pragma_colors =  ['#ffb900', '#8cab13', '#22ab82', '#242526', '#cc0000',  '#ffcc00', '#ffffff']\n",
        "\n",
        "fig, ax = plt.subplots()\n",
        "rects1 = ax.bar(x - width, match_metrics_recall_title , width, label='With multi-match query title field', color=pragma_colors[0])\n",
        "rects2 = ax.bar(x, bert_title_metrics_recall, width, label='With BERT on title field', color=pragma_colors[1])\n",
        "rects3 = ax.bar(x + width, rerank_bert_title_metrics_recall, width, label='With re-ranking on title field', color=pragma_colors[3])\n",
        "\n",
        "# Add some text for labels, title and custom x-axis tick labels, etc.\n",
        "ax.set_ylabel('metric scores')\n",
        "ax.set_title('Recall scores on title field by corpus')\n",
        "ax.set_xticks(x)\n",
        "ax.set_xticklabels(labels)\n",
        "ax.legend()\n",
        "\n",
        "def autolabel(rects):\n",
        "    \"\"\"Attach a text label above each bar in *rects*, displaying its height.\"\"\"\n",
        "    for rect in rects:\n",
        "        height = rect.get_height()\n",
        "        ax.annotate('{}'.format(height),\n",
        "                    xy=(rect.get_x() + rect.get_width() / 2, height),\n",
        "                    xytext=(0, 3),  # 3 points vertical offset\n",
        "                    textcoords=\"offset points\",\n",
        "                    ha='center', va='bottom')\n",
        "\n",
        "autolabel(rects1)\n",
        "autolabel(rects2)\n",
        "autolabel(rects3)\n",
        "\n",
        "\n",
        "fig.tight_layout()\n",
        "fig.set_figwidth(16)\n",
        "fig.set_figheight(10)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W7lZN72V1qRo"
      },
      "source": [
        "**Recall on the 'text' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 531
        },
        "id": "lg18KuWF1qRo",
        "outputId": "c1ff75a3-c295-4997-c526-46e29e4ff0cf"
      },
      "source": [
        "import matplotlib\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "match_metrics_recall_text = []\n",
        "match_metrics_recall_text.append(round(adi_res_match_recall_text['metric_score'], 3))\n",
        "match_metrics_recall_text.append(round(cacm_res_match_recall_text['metric_score'], 3))\n",
        "match_metrics_recall_text.append(round(cisi_res_match_recall_text['metric_score'], 3))\n",
        "match_metrics_recall_text.append(round(cran_res_match_recall_text['metric_score'], 3))\n",
        "match_metrics_recall_text.append(round(lisa_res_match_recall_text['metric_score'], 3))\n",
        "match_metrics_recall_text.append(round(med_res_match_recall_text['metric_score'], 3))\n",
        "match_metrics_recall_text.append(round(npl_res_match_recall_text['metric_score'], 3))\n",
        "match_metrics_recall_text.append(round(time_res_match_recall_text['metric_score'], 3))\n",
        "\n",
        "bert_text_metrics_recall = []\n",
        "bert_text_metrics_recall.append(round(adi_res_bert_text_recall['metric_score'], 3))\n",
        "bert_text_metrics_recall.append(round(cacm_res_bert_text_recall['metric_score'], 3))\n",
        "bert_text_metrics_recall.append(round(cisi_res_bert_text_recall['metric_score'], 3))\n",
        "bert_text_metrics_recall.append(round(cran_res_bert_text_recall['metric_score'], 3))\n",
        "bert_text_metrics_recall.append(round(lisa_res_bert_text_recall['metric_score'], 3))\n",
        "bert_text_metrics_recall.append(round(med_res_bert_text_recall['metric_score'], 3))\n",
        "bert_text_metrics_recall.append(round(npl_res_bert_text_recall['metric_score'], 3))\n",
        "bert_text_metrics_recall.append(round(time_res_bert_text_recall['metric_score'], 3))\n",
        "\n",
        "rerank_bert_text_metrics_recall = []\n",
        "rerank_bert_text_metrics_recall.append(round(adi_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(cacm_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(cisi_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(cran_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(lisa_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(med_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(npl_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(time_res_rerank_text_recall['metric_score'], 3))\n",
        "\n",
        "\n",
        "labels = ['ADI','CACM', 'CISI','Cranfield', 'LISA', 'Medline', 'NPL', 'Time']\n",
        "\n",
        "x = np.arange(len(labels))*2.5  # the label locations\n",
        "\n",
        "width = 0.6  # the width of the bars\n",
        "\n",
        "# orange, green, turquoise, black, red, yellow, white\n",
        "pragma_colors =  ['#ffb900', '#8cab13', '#22ab82', '#242526', '#cc0000',  '#ffcc00', '#ffffff']\n",
        "\n",
        "fig, ax = plt.subplots()\n",
        "rects1 = ax.bar(x - width, match_metrics_recall_text , width, label='With multi-match query text field', color=pragma_colors[0])\n",
        "rects2 = ax.bar(x, bert_text_metrics_recall, width, label='With BERT on text field', color=pragma_colors[1])\n",
        "rects3 = ax.bar(x + width, rerank_bert_text_metrics_recall, width, label='With re-ranking on text field', color=pragma_colors[3])\n",
        "\n",
        "# Add some text for labels, title and custom x-axis tick labels, etc.\n",
        "ax.set_ylabel('metric scores')\n",
        "ax.set_title('Recall scores on text field by corpus')\n",
        "ax.set_xticks(x)\n",
        "ax.set_xticklabels(labels)\n",
        "ax.legend()\n",
        "\n",
        "def autolabel(rects):\n",
        "    \"\"\"Attach a text label above each bar in *rects*, displaying its height.\"\"\"\n",
        "    for rect in rects:\n",
        "        height = rect.get_height()\n",
        "        ax.annotate('{}'.format(height),\n",
        "                    xy=(rect.get_x() + rect.get_width() / 2, height),\n",
        "                    xytext=(0, 3),  # 3 points vertical offset\n",
        "                    textcoords=\"offset points\",\n",
        "                    ha='center', va='bottom')\n",
        "\n",
        "autolabel(rects1)\n",
        "autolabel(rects2)\n",
        "autolabel(rects3)\n",
        "\n",
        "\n",
        "fig.tight_layout()\n",
        "fig.set_figwidth(16)\n",
        "fig.set_figheight(8)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x576 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9tQ8B-fg1qRp"
      },
      "source": [
        "**Recall on a complex re-ranking query**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 647
        },
        "id": "pwQwjB-71qRp",
        "outputId": "8b0aa4c6-b84b-4410-c3d5-b66ea14b7bb8"
      },
      "source": [
        "import matplotlib\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "match_metrics_recall_text_title = []\n",
        "match_metrics_recall_text_title.append(round(adi_res_match_recall_text_title['metric_score'], 3))\n",
        "match_metrics_recall_text_title.append(round(cacm_res_match_recall_text_title['metric_score'], 3))\n",
        "match_metrics_recall_text_title.append(round(cisi_res_match_recall_text_title['metric_score'], 3))\n",
        "match_metrics_recall_text_title.append(round(cran_res_match_recall_text_title['metric_score'], 3))\n",
        "match_metrics_recall_text_title.append(round(lisa_res_match_recall_text_title['metric_score'], 3))\n",
        "\n",
        "complex_rerank_bert_metrics_recall = []\n",
        "complex_rerank_bert_metrics_recall.append(round(adi_res_complex_rerank_recall['metric_score'], 3))\n",
        "complex_rerank_bert_metrics_recall.append(round(cacm_res_complex_rerank_recall['metric_score'], 3))\n",
        "complex_rerank_bert_metrics_recall.append(round(cisi_res_complex_rerank_recall['metric_score'], 3))\n",
        "complex_rerank_bert_metrics_recall.append(round(cran_res_complex_rerank_recall['metric_score'], 3))\n",
        "complex_rerank_bert_metrics_recall.append(round(lisa_res_complex_rerank_recall['metric_score'], 3))\n",
        "\n",
        "rerank_bert_text_metrics_recall = []\n",
        "rerank_bert_text_metrics_recall.append(round(adi_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(cacm_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(cisi_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(cran_res_rerank_text_recall['metric_score'], 3))\n",
        "rerank_bert_text_metrics_recall.append(round(lisa_res_rerank_text_recall['metric_score'], 3))\n",
        "\n",
        "rerank_bert_title_metrics_recall = []\n",
        "rerank_bert_title_metrics_recall.append(round(adi_res_rerank_title_recall['metric_score'], 3))\n",
        "rerank_bert_title_metrics_recall.append(round(cacm_res_rerank_title_recall['metric_score'], 3))\n",
        "rerank_bert_title_metrics_recall.append(round(cisi_res_rerank_title_recall['metric_score'], 3))\n",
        "rerank_bert_title_metrics_recall.append(round(cran_res_rerank_title_recall['metric_score'], 3))\n",
        "rerank_bert_title_metrics_recall.append(round(lisa_res_rerank_title_recall['metric_score'], 3))\n",
        "\n",
        "\n",
        "labels = ['ADI','CACM', 'CISI','Cranfield', 'LISA']\n",
        "\n",
        "x = np.arange(len(labels))*2.5  # the label locations\n",
        "\n",
        "width = 0.5  # the width of the bars\n",
        "\n",
        "# orange, green, turquoise, black, red, yellow, white\n",
        "pragma_colors =  ['#ffb900', '#8cab13', '#22ab82', '#242526', '#cc0000',  '#ffcc00', '#ffffff']\n",
        "\n",
        "fig, ax = plt.subplots()\n",
        "rects1 = ax.bar(x - 2*width, match_metrics_recall_text_title , width, label='With multi-match query on title and text field', color=pragma_colors[0])\n",
        "rects2 = ax.bar(x - width, complex_rerank_bert_metrics_recall , width, label='With complex re-ranking', color=pragma_colors[3])\n",
        "rects3 = ax.bar(x, rerank_bert_text_metrics_recall, width, label='With re-ranking on text field', color=pragma_colors[1])\n",
        "rects4 = ax.bar(x + width, rerank_bert_title_metrics_recall, width, label='With re-ranking on title field', color=pragma_colors[2])\n",
        "\n",
        "# Add some text for labels, title and custom x-axis tick labels, etc.\n",
        "ax.set_ylabel('metric scores')\n",
        "ax.set_title('Complex re-ranking recall scores by corpus')\n",
        "ax.set_xticks(x)\n",
        "ax.set_xticklabels(labels)\n",
        "ax.legend()\n",
        "\n",
        "def autolabel(rects):\n",
        "    \"\"\"Attach a text label above each bar in *rects*, displaying its height.\"\"\"\n",
        "    for rect in rects:\n",
        "        height = rect.get_height()\n",
        "        ax.annotate('{}'.format(height),\n",
        "                    xy=(rect.get_x() + rect.get_width() / 2, height),\n",
        "                    xytext=(0, 3),  # 3 points vertical offset\n",
        "                    textcoords=\"offset points\",\n",
        "                    ha='center', va='bottom')\n",
        "\n",
        "autolabel(rects1)\n",
        "autolabel(rects2)\n",
        "autolabel(rects3)\n",
        "autolabel(rects4)\n",
        "\n",
        "fig.tight_layout()\n",
        "fig.set_figwidth(16)\n",
        "fig.set_figheight(10)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "rhLH9vIv1qRr"
      },
      "source": [
        "### Precision"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dkrvGhj61qRr"
      },
      "source": [
        "**Precision on the 'title' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 647
        },
        "id": "ORIzDR811qRr",
        "outputId": "dee04159-d966-4967-9610-e45035860799"
      },
      "source": [
        "import matplotlib\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "match_metrics_precision_title = []\n",
        "match_metrics_precision_title.append(round(adi_res_match_precision_title['metric_score'], 3))\n",
        "match_metrics_precision_title.append(round(cacm_res_match_precision_title['metric_score'], 3))\n",
        "match_metrics_precision_title.append(round(cisi_res_match_precision_title['metric_score'], 3))\n",
        "match_metrics_precision_title.append(round(cran_res_match_precision_title['metric_score'], 3))\n",
        "match_metrics_precision_title.append(round(lisa_res_match_precision_title['metric_score'], 3))\n",
        "\n",
        "bert_title_metrics_precision = []\n",
        "bert_title_metrics_precision.append(round(adi_res_bert_title_precision['metric_score'], 3))\n",
        "bert_title_metrics_precision.append(round(cacm_res_bert_title_precision['metric_score'], 3))\n",
        "bert_title_metrics_precision.append(round(cisi_res_bert_title_precision['metric_score'], 3))\n",
        "bert_title_metrics_precision.append(round(cran_res_bert_title_precision['metric_score'], 3))\n",
        "bert_title_metrics_precision.append(round(lisa_res_bert_title_precision['metric_score'], 3))\n",
        "\n",
        "rerank_bert_title_metrics_precision = []\n",
        "rerank_bert_title_metrics_precision.append(round(adi_res_rerank_title_precision['metric_score'], 3))\n",
        "rerank_bert_title_metrics_precision.append(round(cacm_res_rerank_title_precision['metric_score'], 3))\n",
        "rerank_bert_title_metrics_precision.append(round(cisi_res_rerank_title_precision['metric_score'], 3))\n",
        "rerank_bert_title_metrics_precision.append(round(cran_res_rerank_title_precision['metric_score'], 3))\n",
        "rerank_bert_title_metrics_precision.append(round(lisa_res_rerank_title_precision['metric_score'], 3))\n",
        "\n",
        "labels = ['ADI','CACM', 'CISI','Cranfield', 'LISA']\n",
        "\n",
        "x = np.arange(len(labels))*2.5  # the label locations\n",
        "\n",
        "width = 0.5  # the width of the bars\n",
        "\n",
        "# orange, green, turquoise, black, red, yellow, white\n",
        "pragma_colors =  ['#ffb900', '#8cab13', '#22ab82', '#242526', '#cc0000',  '#ffcc00', '#ffffff']\n",
        "\n",
        "fig, ax = plt.subplots()\n",
        "rects1 = ax.bar(x - width, match_metrics_precision_title , width, label='With multi-match query on title', color=pragma_colors[0])\n",
        "rects2 = ax.bar(x, bert_title_metrics_precision, width, label='With BERT on title field', color=pragma_colors[1])\n",
        "rects3 = ax.bar(x + width, rerank_bert_title_metrics_precision, width, label='With re-ranking on title field', color=pragma_colors[3])\n",
        "\n",
        "# Add some text for labels, title and custom x-axis tick labels, etc.\n",
        "ax.set_ylabel('metric scores')\n",
        "ax.set_title('Precision scores on title field by corpus')\n",
        "ax.set_xticks(x)\n",
        "ax.set_xticklabels(labels)\n",
        "ax.legend()\n",
        "\n",
        "def autolabel(rects):\n",
        "    \"\"\"Attach a text label above each bar in *rects*, displaying its height.\"\"\"\n",
        "    for rect in rects:\n",
        "        height = rect.get_height()\n",
        "        ax.annotate('{}'.format(height),\n",
        "                    xy=(rect.get_x() + rect.get_width() / 2, height),\n",
        "                    xytext=(0, 3),  # 3 points vertical offset\n",
        "                    textcoords=\"offset points\",\n",
        "                    ha='center', va='bottom')\n",
        "\n",
        "autolabel(rects1)\n",
        "autolabel(rects2)\n",
        "autolabel(rects3)\n",
        "\n",
        "fig.tight_layout()\n",
        "fig.set_figwidth(16)\n",
        "fig.set_figheight(10)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAABBcAAAJ2CAYAAAAExGr0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOzdeXxU1f3/8feHEEAEggH0J0sBQRBMQggQNIigIqGKcalW1Mpi0aJFXLG4FNG60C+4lEXUrwtWUVG0mC9iVagoOFEMGJBNEBvZ2krYQbaQ8/tjbqaTjQzcDDDyej4eeZC599xzPndmQOc9555rzjkBAAAAAAAcrmpHuwAAAAAAABDbCBcAAAAAAIAvhAsAAAAAAMAXwgUAAAAAAOAL4QIAAAAAAPCFcAEAAAAAAPhCuAAAOO6Z2XVm9lEE7Z41sz8eiZqOR2Z2n5m9cJD9A81s3mH2fYqZfWZmO8zsicrGKnXsHDMbXMG+FmbmzKx6BfvzzazX4dQMAEAsKfc/hAAAHCvMLF/SKZIOSNol6QNJQ51zO6tqDOfcFElTImg3pKrGPN6ZWU9JrznnmhZvc849Fra/haR/Sop3zhVWwZA3SSqQVM8556qgPwAAEIaZCwCAWHCJc66OpDRJnSU9ULpBRd8cg+fG01zSsuM9WOC9AACIFsIFAEDMcM6tV3DmQpIkedPRf29mqySt8rb1NbM8M9tqZgEzSyk+3syamdm7ZrbRzDaZ2QRve2i6vQU9ZWY/mtl2M/vGzIrHm2xmj4T1d6OZfWdmm80s28wah+1zZjbEzFZ5tUw0MyvvvMws3cxyvfH+Y2ZPhu07xzuPrWa21swGetsTzOyv3rn8YGYPmFm1sPP53DuPTZJGmVlNMxtrZmu8MZ41sxO89g3NbIY3xmYzm1vcVzm1ZpjZV2a2zfszI2zfHDP7kzf2DjP7yMwaltPHid7r2NjMdno/jc1slJm95jX7zPtzq7f/7HL6OcPMPvZq/tbMfl1BzZMlDZB0j9dXr1JjyczOCnueF3kzK8rrK857HgvM7HtJF5fXrpQuZrbMzLaY2ctmVsvra4mZXRLWd7zXb8cKxr7Ue29vN7PVZtbH297Ye/9t9t6PN4YdM8rMppnZa2a2XdLAsG1TvddpoZl1CDvGmVnr8Oev+H1/KO8VAMDxhf8YAABihpk1k3SRpK/DNl8mqauk9t6Hspck/U5SA0nPScr2PljHSZoh6QdJLSQ1kfRmOcP0lnSupDaSEiT9WtKmcmo5X9Lj3v5TvX5L99dXUhdJKV67zApO7S+S/uKcqyeplaS3vDGaK/ghfLykRpJSJeV5x4z36jtNUg9J/SUNCuuzq6TvFbyk5FFJo71zSpXU2jv/kV7buySt88Y4RdJ9ksp8w29miZLelzROwef3SUnvm1mDsGbXenWcLKmGpLtL9+Oc2yXpl5I2OOfqeD8bSjU71/uzvrc/p1QtJ0r6WNLr3lj9JD1jZu3LGW+ggpe9/I/X16xSfTXxzusRSYleze+YWaPSfUm6UcHXtaOCs2iuLKdNadcp+Nq3UvA1KJ5581dJvwlrd5GkfznnvlYpZpbutR8uqb6Cz0++t/tNBV+/xl49j3nvz2KXSprmHTclbNvb3vm+Lmm6mcVHcC4RvVcAAMcfwgUAQCyYbmZbJc2T9Kmkx8L2Pe6c2+yc263gdfXPOee+dM4dcM69ImmvpLMkpSv44Wu4c26Xc26Pc668xQH3S6or6QxJ5pxb7pz7VzntrpP0knNuoXNur6R7JZ1twbUCio12zm11zq2R9ImCH+zLs19SazNr6Jzb6Zz7wtt+raRZzrk3nHP7nXObnHN5XlDST9K9zrkdzrl8SU9Iuj6szw3OufHeegV7vOfmDu+52uE9h/3Cxj9VUnNvnLkVXD5wsaRVzrlXnXOFzrk3JK2QdElYm5edcyu91+Otg5yzX30l5TvnXvZq+VrSO5KuOoy+fiNppnNupnOuyDn3saRcBT/sl/ZrSU8759Y65zYrGDBVZkJY+0clXeNtf03SRWZWz3t8vaRXK+jjtwq+3z72alzvnFvhBW7dJP3Be0/nSXpBwbCpWI5zbrp33G5v2wLn3DTn3H4FQ6JaCv49qUyk7xUAwHGGcAEAEAsuc87Vd841d87dEvYBSZLWhv3eXNJd3pTtrV4g0UzBUKGZpB8qWxzQOfcPSRMkTZT0o5k9H/bhL1xjBWcrFB+3U8EZDk3C2vw77PefJNWpYNjfKviN9grvUoO+3vZmklaX076hpPjw8b3fw8cOf14aSaotaUHY8/J3b7skjZH0naSPzOx7MxtRQZ0lzrmCcSM9Z7+aS+pa6rW+TtL/O8y+rirV1zkKfogurbFKPreln4/ylG7fWJK82RqfS/qVmdVXcDZHRQuLVvReaCypODAKH6Oi90KZbc65Iv135kNlIn2vAACOM4QLAIBYF/6t6VpJj3pBRPFPbe8b9rWSfmERLGjnnBvnnOskqb2CH/qHl9Nsg4IfSiWFpuk3kLT+kE/AuVXOuWsUnN7/Z0nTvP7WKjiVvrQCBb9Bbh627Relxnal2u+WdGbY85LgLZIpb/bDXc650yRlSbrTzC4oZ9wS51zBuJGq7NvuyvavlfRpqde6jnPu5sOoZa2kV0v1daJzbnQ5bf+l4Af9Yr+IoP/S7cMvAXlFwZkTVyk4w6Ci57Ki98IGSYlmVrfUGBW9F8rU5K2Z0DSsrp8UDKOKhQKbQ3ivAACOM4QLAICfk/+VNMTMulrQiWZ2sffBa76CHwxHe9trmVm30h2YWRfv+HgFb325R1JROWO9IWmQmaWaWU0FLzP40rtE4ZCY2W/MrJH3DfJWb3ORgt9i9zKzX5tZdTNrYGapzrkDCl5y8KiZ1fXWZrhTwWn2ZXj9/q+kp8zsZG/MJmaW6f3e18xam5lJ2qbgbT/LO+eZktqY2bVePVcrGMDMONRzlvQfSQ3MLKGC/Ru9Gk6rYP8Mr5brvYUQ473Xrt1h1PKapEvMLNOCCzbWMrOeZta0nLZvSRpmZk3N7CRJkXxz/3uvfaKk+yVNDds3XcG7oNym4JoKFXlRwffbBWZWzXv9znDOrZUUkPS4V3eKgjNhyn0vhOlkZld4YdvtCl4+VHw5Tp6ka73noo+Ca3pIOqT3CgDgOEO4AAD42XDO5Sq44N4ESVsUnL490Nt3QMG1AVpLWqPgNPCry+mmnoIfxLcoOL18k4JTwUuPNUvSHxW8zv9fCn6r3K90uwj1kbTUzHYquLhjP+fcbm+thosUXERvs4If+opX9b9VwfDjewXXonhdwcUsK/IHBZ+PLyx414BZktp6+073Hu+UlCPpGefcJ+Wc8yYF1zq4S8Hn5R5JfZ1zBYd6ws65FQoGNN97lyI0LrX/JwXXJ/jc239Wqf07FFx8s5+C37j/W8FZHzUPo5a1Ci5weJ+CocZaBWerlPf/Sf8r6UNJiyQtlPRuBEO8LukjBV+r1QouHFk89m4F30MtD9aXc26+ggtlPqXgh/pP9d9ZJNcouEjpBkl/k/Rg6UUry/Gegu//LQqu9XCFt/6CFAw6LlEw6LpOwQCkWETvFQDA8cdYgwcAAODoMbORkto4535TaeOqGW+UpNZHajwAwPGh0utOAQAAEB3epRK/Vck7fQAAEHO4LAIAAOAoMLMbFbwE4wPn3GdHux4AAPzgsggAAAAAAOALMxcAAAAAAIAvMbfmQsOGDV2LFi2OdhkAAAAAABxXFixYUOCca1TevpgLF1q0aKHc3NyjXQYAAAAAAMcVM/uhon1cFgEAAAAAAHwhXAAAAAAAAL4QLgAAAAAAAF9ibs0FAAAAAIi2/fv3a926ddqzZ8/RLgU44mrVqqWmTZsqPj4+4mMIFwAAAACglHXr1qlu3bpq0aKFzOxolwMcMc45bdq0SevWrVPLli0jPo7LIgAAAACglD179qhBgwYECzjumJkaNGhwyLN2CBcAAAAAoBwECzheHc57n3ABAAAAAAD4QrgAAAAAAJX5yKr2pxJ33HGHnn766dDjzMxMDR48OPT4rrvu0pNPPqns7GyNHj1akjR9+nQtW7Ys1KZnz57Kzc2twiehrPz8fCUlJUmS8vLyNHPmzNC+8NqqeszXX3+90nYtWrRQQUFBlY9/LHn66af1008/hR5fdNFF2rp1q7Zu3apnnnkmtD38dYoWwgUAAAAAOMZ069ZNgUBAklRUVKSCggItXbo0tD8QCCgjI0NZWVkaMWKEpLLhwpFWOlwIr60qRRouHE2FhYVHZJzS4cLMmTNVv379MuHCkUC4AAAAAADHmIyMDOXk5EiSli5dqqSkJNWtW1dbtmzR3r17tXz5cqWlpWny5MkaOnSoAoGAsrOzNXz4cKWmpmr16tWSpLffflvp6elq06aN5s6dW2acOXPmqEePHrr00kt12mmnacSIEZoyZYrS09OVnJwc6mfgwIGaNm1a6Lg6deqU6Gffvn0aOXKkpk6dqtTUVE2dOjVUW3nq1Kmj4cOH68wzz1SvXr00f/589ezZU6eddpqys7MlBUOE7t27Ky0tTWlpaaGwZcSIEZo7d65SU1P11FNP6cCBA7r77ruVlJSklJQUjR8/PjTO+PHjlZaWpuTkZK1YsaJMHbt371a/fv3Url07XX755eratWtotkf4OU6bNk0DBw6UJG3cuFG/+tWv1KVLF3Xp0kWff/65JGnUqFG6/vrr1a1bN11//fU699xzlZeXF+rjnHPO0aJFi0qMv2fPHg0aNEjJycnq2LGjPvnkE0nS5MmTdcUVV6hPnz46/fTTdc8995Spfdy4cdqwYYPOO+88nXfeeZL+O1tjxIgRWr16tVJTUzV8+PASxx04cEDDhw9Xly5dlJKSoueee67c1+hQcStKAAAAADjGNG7cWNWrV9eaNWsUCAR09tlna/369crJyVFCQoKSk5NVo0aNUPviWQx9+/bVlVdeGdpeWFio+fPna+bMmXrooYc0a9asMmMtWrRIy5cvV2Jiok477TQNHjxY8+fP11/+8heNHz++xOUZFalRo4Yefvhh5ebmasKECZKCH5ArsmvXLp1//vkaM2aMLr/8cj3wwAP6+OOPtWzZMg0YMEBZWVk6+eST9fHHH6tWrVpatWqVrrnmGuXm5mr06NEaO3asZsyYIUmaNGmS8vPzlZeXp+rVq2vz5s2hcRo2bKiFCxfqmWee0dixY/XCCy+UqGPSpEmqXbu2li9frsWLFystLa3Sc73tttt0xx136JxzztGaNWuUmZmp5cuXS5KWLVumefPm6YQTTtArr7yiyZMn6+mnn9bKlSu1Z88edejQoURfEydOlJnpm2++0YoVK9S7d2+tXLlSUnAmyNdff62aNWuqbdu2uvXWW9WsWbPQscOGDdOTTz6pTz75RA0bNizR7+jRo7VkyZJQuJGfnx/a9+KLLyohIUFfffWV9u7dq27duql3796HdNvJ8hAuAAAAAMAxKCMjQ4FAQIFAQHfeeafWr1+vQCCghIQEdevWLaI+rrjiCklSp06dSnzADNelSxedeuqpkqRWrVqpd+/ekqTk5OTQN+lVrUaNGurTp09onJo1ayo+Pl7JycmhOvfv36+hQ4cqLy9PcXFxoQ/dpc2aNUtDhgxR9erBj7eJiYmhfeHn/+6775Y59rPPPtOwYcMkSSkpKUpJSam09lmzZpW4/GT79u3auXOnpOClICeccIIk6aqrrtKf/vQnjRkzRi+99FJo5kO4efPm6dZbb5UknXHGGWrevHnoPC+44AIlJCRIktq3b68ffvihRLhwuD766CMtXrw4NBNl27ZtWrVqFeECAAAAAPwcFa+78M033ygpKUnNmjXTE088oXr16mnQoEER9VGzZk1JUlxcXIXrABS3kaRq1aqFHlerVi10TPXq1VVUVCQpuAbEvn37Ij6PAwcOqFOnTpKCH74ffvhhxcfHh253WNGYTz31lE455RQtWrRIRUVFqlWrVsRjlj63g51/RcJvx7hnz57Q70VFRfriiy/KrefEE08M/V67dm1deOGFeu+99/TWW29pwYIFh1W7dHj1V8Q5p/HjxyszM7NK+ivGmgs4qL///e9q27atWrdufdCVXt955x2ZWej6pPz8fJ1wwglKTU1VamqqhgwZEmo7depUpaSk6Mwzz9Qf/vCHqJ8DAAAAEIsyMjI0Y8YMJSYmKi4uTomJidq6datycnKUkZFRpn3dunW1Y8eOqNTSokWL0Ifj7Oxs7d+/P+Lx4+LilJeXp7y8PD388MMRj7lt2zadeuqpqlatml599VUdOHCg3HEuvPBCPffcc6EP3+GXRVTm3HPPDS0OuWTJEi1evDi075RTTtHy5ctVVFSkv/3tb6HtvXv3LrGuQ/i6CqUNHjxYw4YNU5cuXXTSSSeV2d+9e3dNmTJFkrRy5UqtWbNGbdu2jbj+ip7zg70XMjMzNWnSpNBruHLlSu3atSviMStCuIAKHThwQL///e/1wQcfaNmyZXrjjTfKXX12x44d+stf/qKuXbuW2N6qVavQPyLPPvusJGnTpk0aPny4Zs+eraVLl+rf//63Zs+efUTOBwAAADhsvV3V/kQgOTlZBQUFOuuss0psS0hIKHONvST169dPY8aMUceOHUMLMVaVG2+8UZ9++qk6dOignJycEt/QFzvvvPO0bNmy0IKOft1yyy165ZVX1KFDB61YsSI0ZkpKiuLi4tShQwc99dRTGjx4sH7xi18oJSVFHTp0OKQ7Sdx8883auXOn2rVrp5EjR4ZmWEjBdQv69u2rjIyM0GUjUnAhxdzcXKWkpKh9+/ahzzrl6dSp00Fnmtxyyy0qKipScnKyrr76ak2ePLnEjIXK3HTTTerTp09oQcdiDRo0ULdu3ZSUlFRmQcfBgwerffv2SktLU1JSkn73u99VyawIcy6yN/axonPnzi7a92pFUE5OjkaNGqUPP/xQkvT4449Lku69994S7W6//XZdeOGFGjNmjMaOHavOnTsrPz9fffv21ZIlS0q0/eqrrzRixIhQoPDqq68qJyfniN8mBQAAADiY5cuXq127dke7DBxhPXv2DH2mqQobNmxQz549tWLFClWrFlvf7Zf3d8DMFjjnyn1yYuvscEStX7++xIIhTZs21fr160u0WbhwodauXauLL764zPH//Oc/1bFjR/Xo0SN025vWrVvr22+/VX5+vgoLCzV9+nStXbs2uicCAAAAAEfYX//6V3Xt2lWPPvpozAULh4MFHXHYioqKdOedd5Z7i5lTTz1Va9asUYMGDbRgwQJddtllWrp0qU466SRNmjRJV199tapVq6aMjIwqn7IFAAAAAIdjzpw5VdZX//791b9//yrr71j3849PcNiaNGlSYlbBunXr1KRJk9DjHTt2aMmSJerZs6datGihL774QllZWcrNzVXNmjXVoEEDScHrjFq1ahW6pcoll1yiL7/8Ujk5OWrbtq3atGlzZE8MAAAAAFClCBdQoS5dumjVqlX65z//qX379unNN99UVlZWaH9CQoIKCgqUn5+v/Px8nXXWWcrOzlbnzp21cePG0Gqu33//vVatWqXTTjtNkvTjjz9KkrZs2aJnnnlGgwcPPvInBwAAAACoMlwWgQpVr15dEyZMUGZmpg4cOKAbbrhBZ555pkaOHKnOnTuXCBpK++yzzzRy5EjFx8erWrVqevbZZ5WYmChJuu2227Ro0SJJ0siRI5m5AAAAAAAxjrtFAAAAAEAp3C0Cx7tDvVsEMxcAAAAAoBJPTGtYpf3ddWXBQfffcccdat68uW6//XZJUmZmppo1a6YXXnghePxdd6lJkyZq3bq1li1bphEjRmj69Olq06aN2rdvLymy2yrm5+erXbt2atu2rZxzOvHEE/Xyyy+rbdu2mjNnji699FK1bNky1H7s2LHq1auX4uLilJycrMLCQrVs2VKvvvqqMjMztXfvXm3evFm7d+8Ordc2ffp0tWjRws/TFZKXl6cNGzbooosukiRlZ2dXeP4DBw5U3759deWVV0bU98aNG9W3b1/t27dP48aN0+OPP67XX39d9evXr/CYFi1aKDc3Vw0blnx/jBo1SnXq1NHdd999mGcae1hzAQAAAACOMd26dVMgEJAUvEtbQUGBli5dGtofCASUkZGhrKwsjRgxQlLwQ/yyZcsOeaxWrVopLy9PixYt0oABA/TYY4+F9nXv3l15eXmhn169ekmSTjjhBOXl5WnJkiVKTEzUxIkT9eWXXyovL08PP/ywrr766tAxVRUsSMFwYebMmaHHVXH+xWbPnq3k5GR9/fXX6t69u2bOnHnQYAElES4AAAAAwDEmIyNDOTk5kqSlS5cqKSlJdevW1ZYtW7R3714tX75caWlpmjx5soYOHapAIKDs7GwNHz5cqampodu9v/3220pPT1ebNm00d+7cSsfdvn27TjrppEOq9eyzz9b69esjbj979mx17NhRycnJuuGGG7R3715JwVkADz74oNLS0pScnKwVK1aUOG7fvn0aOXKkpk6dqtTUVE2dOrXS8y+2YMEC9ejRQ506dVJmZqb+9a9/ldifl5ene+65R++9955SU1O1e/dutWjRQgUFwRkmr732mtLT05Wamqrf/e53ocXrwz366KNq06aNzjnnHH377bcRPx8/F1wWgSqf4lWVKpsuBgAAAPwcNW7cWNWrV9eaNWsUCARCH+BzcnKUkJCg5ORk1ahRI9S+eBZD6csACgsLNX/+fM2cOVMPPfSQZs2aVWas1atXKzU1VTt27NBPP/2kL7/8MrRv7ty5Sk1NDT1+55131KpVq9DjAwcOaPbs2frtb38b0Xnt2bNHAwcO1OzZs9WmTRv1799fkyZNCl3+0bBhQy1cuFDPPPOMxo4dG7oMRJJq1Kihhx9+WLm5uZowYYIkafLkyQc9f0nav3+/br31Vr333ntq1KiRpk6dqvvvv18vvfRSqE1qamqZvostX75cU6dO1eeff674+HjdcsstmjJlivr37x9qs2DBAr355pvKy8tTYWGh0tLS1KlTp4iek58LwgUAAAAAOAZlZGQoEAgoEAjozjvv1Pr16xUIBJSQkKBu3bpF1McVV1whSerUqZPy8/PLbVN8WYQkTZ06VTfddJP+/ve/SwpeFjFjxowyx+zevVupqalav3692rVrpwsvvDCier799lu1bNkydMe4AQMGaOLEiaFwIbzed999N6I+IxlzyZIloRoPHDigU089NeLjZ8+erQULFqhLly6Sgud+8sknl2gzd+5cXX755apdu7YkHfTOej9XhAsAAAAAcAwqXnfhm2++UVJSkpo1a6YnnnhC9erV06BBgyLqo2bNmpKkuLg4FRYWVto+Kysror6L11z46aeflJmZqYkTJ2rYsGER1VSV9UbCOaczzzwzdJnJ4Rw/YMAAPf7441VSz88Vay4AAAAAwDEoIyNDM2bMUGJiouLi4pSYmKitW7cqJydHGRkZZdrXrVtXO3bs8DXmvHnzSlz2UJnatWtr3LhxeuKJJyIKA9q2bav8/Hx99913kqRXX31VPXr0iHi8g51jRfvatm2rjRs3hsKF/fv3l1gcszIXXHCBpk2bph9//FGStHnzZv3www8l2px77rmaPn26du/erR07duj//u//Iu7/54KZCwAAAABQiaOxFlhycrIKCgp07bXXlti2c+fOMrc+lKR+/frpxhtv1Lhx4zRt2rSIxylec8E5pxo1apRY56D0mgsPPPBAmTUNOnbsqJSUFL3xxhu6/vrrDzpWrVq19PLLL+uqq65SYWGhunTpoiFDhkRc63nnnafRo0crNTVV9957b4l9FZ1/jRo1NG3aNA0bNkzbtm1TYWGhbr/9dp155pkRjdm+fXs98sgj6t27t4qKihQfH6+JEyeqefPmoTZpaWm6+uqr1aFDB5188smhSyiOJ+acO9o1HJLOnTu73Nzco13GzwoLOgIAAAAlLV++XO3atTvaZQBHTXl/B8xsgXOuc3ntuSwCAAAAAAD4QrgAAAAAAAB8IVwAAAAAAAC+EC4AAAAAAABfCBcAAAAAAIAvhAsAAAAAAMCX6ke7AAAAAAA41p3epm2V9rdq5bcH3X/HHXeoefPmuv322yVJmZmZatasmV544QVJ0l133aUmTZqodevWWrZsmUaMGKHp06erTZs2at++vSSpZ8+eGjt2rDp3LvfOgUddnTp1tHPnzhLbNmzYoGHDhmnatGlHqar/euyxx3TfffeFHmdkZCgQCCg/P1+BQEDXXnutJGnOnDkaO3asZsyYEXHfw4cP18yZM3XRRRepVatWql27tvr3719h+1GjRqlOnTq6++67S2zPz89X3759tWTJkkM8u6rHzAUAAAAAOMZ069ZNgUBAklRUVKSCggItXbo0tD8QCCgjI0NZWVkaMWKEJGn69OlatmzZYY9ZWFh4RI8rT+PGjY+JYEEKhgvhil+P/Px8vf766776fv7557V48WKNGTNGQ4YMOWiwECsIFwAAAADgGJORkaGcnBxJ0tKlS5WUlKS6detqy5Yt2rt3r5YvX660tDRNnjxZQ4cOVSAQUHZ2toYPH67U1FStXr1akvT2228rPT1dbdq00dy5c8uMM2fOHHXv3l1ZWVlq3769Dhw4oOHDh6tLly5KSUnRc889V259AwcO1JAhQ9S1a1fdc889mj9/vs4++2x17NhRGRkZ+vbb4MyMyZMn64orrlCfPn10+umn65577inTV0FBgc4++2y9//77ys/PV1JSUqXHvvjii2rTpo3S09N14403aujQoWX63bx5sy677DKlpKTorLPO0uLFiyUFZwHccMMN6tmzp0477TSNGzeuzLEjRozQ7t27lZqaquuuu05ScKZF8b65c+cqNTVVTz31VInjdu3apRtuuEHp6enq2LGj3nvvvTJ9Z2VlaefOnerUqZOmTp2qUaNGaezYsZKk1atXq0+fPurUqZO6d++uFStWlDl+wYIF6tChgzp06KCJEyeW8+ocHVwWAQAAAADHmMaNG6t69epas2aNAoGAzj77bK1fv145OTlKSEhQcnKyatSoEWpfPIuhb9++uvLKK0PbCwsLNX/+fM2cOVMPPfSQZs2aVYSqquoAACAASURBVGashQsXasmSJWrZsqWef/55JSQk6KuvvtLevXvVrVs39e7dWy1btixz3Lp16xQIBBQXF6ft27dr7ty5ql69umbNmqX77rtP77zzjiQpLy9PX3/9tWrWrKm2bdvq1ltvVbNmzSRJ//nPf5SVlaVHHnlEF154ofLz80uMUd6xcXFx+tOf/qSFCxeqbt26Ov/889WhQ4cy9T344IPq2LGjpk+frn/84x/q37+/8vLyJEkrVqzQJ598oh07dqht27a6+eabFR8fHzp29OjRmjBhQqh9uNGjR5e4DGLOnDmhfY8++qjOP/98vfTSS9q6davS09PVq1cvnXjiiaE22dnZqlOnTqjvUaNGhfbddNNNevbZZ3X66afryy+/1C233KJ//OMfJcYfNGiQJkyYoHPPPVfDhw8vU9/RQrgAAAAAAMeg4mv8A4GA7rzzTq1fv16BQEAJCQnq1q1bRH1cccUVkqROnTqV+eBeLD09PRQefPTRR1q8eHHo0oRt27Zp1apV5YYLV111leLi4kLtBgwYoFWrVsnMtH///lC7Cy64QAkJCZKk9u3b64cfflCzZs20f/9+XXDBBZo4caJ69OhRbm3lHVtQUKAePXooMTExVMfKlSvLHDtv3rxQwHH++edr06ZN2r59uyTp4osvVs2aNVWzZk2dfPLJ+s9//qOmTZse5JmMzEcffaTs7OzQTIQ9e/ZozZo1ateuXaXH7ty5U4FAQFdddVVo2969e0u02bp1q7Zu3apzzz1XknT99dfrgw8+8F13VSBcAAAAAIBjUPG6C998842SkpLUrFkzPfHEE6pXr54GDRoUUR81a9aUJMXFxVW4NkL4t+rOOY0fP16ZmZkl2tx///16//33JSn0jXv4cX/84x913nnn6W9/+5vy8/PVs2fPMjWUrqN69erq1KmTPvzwwwrDhYqO9Sta/Trn9M4776ht20NfALSoqEj169cvd7ZELGDNBQAAAAA4BmVkZGjGjBlKTExUXFycEhMTtXXrVuXk5CgjI6NM+7p162rHjh2+xszMzNSkSZNCMw9WrlypXbt26dFHH1VeXl6FH3y3bdumJk2aSAqulRAJM9NLL72kFStW6M9//nPENXbp0kWffvqptmzZosLCwtDshNK6d++uKVOmSApeutCwYUPVq1cv4nHi4+NLzMAodrDnOTMzU+PHj5dzTpL09ddfRzxevXr11LJlS7399tuSgkHFokWLSrSpX7++6tevr3nz5klS6PyOBcxcAAAAAIBKVHbryGhITk5WQUFB6JaHxdt27typhg0blmnfr18/3XjjjRo3btxh33Fh8ODBys/PV1pampxzatSokaZPn17pcffcc48GDBigRx55RBdffHHE48XFxemNN95QVlaW6tatq4suuqjSY5o0aaL77rtP6enpSkxM1BlnnBG6dCJc8cKNKSkpql27tl555ZWI65KC6x+kpKQoLS2txIf4lJQUxcXFqUOHDho4cKA6duwY2vfHP/5Rt99+u1JSUlRUVKSWLVse0i0qp0yZoptvvlmPPPKI9u/fr379+pVZT+Lll1/WDTfcIDNT7969D+mcosmKE5VY0blzZ5ebm3u0y/hZeWJa2X+YjhV3XVlwtEsAAADAcWj58uURXSePo2Pnzp2qU6eOCgsLdfnll+uGG27Q5ZdffrTL+lkp7++AmS1wznUurz2XRQAAAAAAYsqoUaOUmpqqpKQktWzZUpdddtnRLum4x2URAAAAAICYUnw3Bhw7mLkAAAAAAOWItUvIgapyOO99wgUAAAAAKKVWrVratGkTAQOOO845bdq0SbVq1Tqk47gsAgAAAABKadq0qdatW6eNGzce7VKAI65WrVpq2rTpIR1DuAAAAAAApcTHx6tly5ZHuwwgZnBZBAAAAAAA8IVwAQAAAAAA+EK4AAAAAAAAfCFcAAAAAAAAvhAuAAAAAAAAXwgXAAAAAACAL4QLAAAAAADAF8IFAAAAAADgC+ECAAAAAADwhXABAAAAAAD4QrgAAAAAAAB8IVwAAAAAAAC+EC4AAAAAAABfCBcAAAAAAIAvhAsAAAAAAMAXwgUAAAAAAOAL4QIAAAAAAPCFcAEAAAAAAPhCuAAAAAAAAHwhXAAAAAAAAL4QLgAAAAAAAF8IFwAAAAAAgC+ECwAAAAAAwBfCBQAAAAAA4AvhAgAAAAAA8IVwAQAAAAAA+EK4AAAAAAAAfCFcAAAAAAAAvhAuAAAAAAAAXwgXAAAAAACAL4QLAAAAAADAF8IFAAAAAADgC+ECAAAAAADwhXABAAAAAAD4QrgAAAAAAAB8IVwAAAAAAAC+EC4AAAAAAABfCBcAAAAAAIAvhAsAAAAAAMAXwgUAAAAAAOAL4QIAAAAAAPAlquGCmfUxs2/N7DszG3GQdr8yM2dmnaNZDwAAAAAAqHpRCxfMLE7SREm/lNRe0jVm1r6cdnUl3Sbpy2jVAgAAAAAAoieaMxfSJX3nnPveObdP0puSLi2n3Z8k/VnSnijWAgAAAAAAoiSa4UITSWvDHq/ztoWYWZqkZs6596NYBwAAAAAAiKKjtqCjmVWT9KSkuyJoe5OZ5ZpZ7saNG6NfHAAAAAAAiFg0w4X1kpqFPW7qbStWV1KSpDlmli/pLEnZ5S3q6Jx73jnX2TnXuVGjRlEsGQAAAAAAHKpohgtfSTrdzFqaWQ1J/SRlF+90zm1zzjV0zrVwzrWQ9IWkLOdcbhRrAgAAAAAAVSxq4YJzrlDSUEkfSlou6S3n3FIze9jMsqI1LgAAAAAAOLKqR7Nz59xMSTNLbRtZQdue0awFAAAAAABEx1Fb0BEAAAAAAPw8EC4AAAAAAABfCBcAAAAAAIAvhAsAAAAAAMAXwgUAAAAAAOAL4QIAAAAAAPCFcAEAAAAAAPhCuAAAAAAAAHwhXAAAAAAAAL4QLgAAAAAAAF8IFwAAAAAAgC+ECwAAAAAAwBfCBQAAAAAA4AvhAgAAAAAA8IVwAQAAAAAA+EK4AAAAAAAAfCFcAAAAAAAAvhAuAAAAAAAAXwgXAAAAAACAL4QLAAAAAADAF8IFAAAAAADgC+ECAAAAAADwhXABAAAAAAD4QrgAAAAAAAB8IVwAAAAAAAC+EC4AAAAAAABfCBcAAAAAAIAvhAsAAAAAAMAXwgUAAAAAAOAL4QIAAAAAAPCFcAEAAAAAAPhCuAAAAAAAAHwhXAAAAAAAAL4QLgAAAAAAAF8IFwAAAAAAgC+ECwAAAAAAwBfCBQAAAAAA4AvhAgAAAAAA8IVwAQAAAAAA+EK4AAAAAAAAfCFcAAAAAAAAvhAuAAAAAAAAXwgXAAAAAACAL4QLAAAAAADAF8IFAAAAAADgC+ECAAAAAADwhXABAAAAAAD4QrgAAAAAAAB8IVwAAAAAAAC+EC4AAAAAAABfCBcAAAAAAIAvhAsAAAAAAMAXwgUAAAAAAOAL4QIAAAAAAPCFcAEAAAAAAPhCuAAAAAAAAHwhXAAAAAAAAL4QLgAAAAAAAF8IFwAAAAAAgC+ECwAAAAAAwBfCBQAAAAAA4AvhAgAAAAAA8IVwAQAAAAAA+EK4AAAAAAAAfCFcAAAAAAAAvhAuAAAAAAAAXwgXAAAAAACAL4QLAAAAAADAF8IFAAAAAADgC+ECAAAAAADwhXABAAAAAAD4QrgAAAAAAAB8IVwAAAAAAAC+EC4AAAAAAABfCBcAAAAAAIAvhAsAAAAAAMAXwgUAAAAAAOAL4QIAAAAAAPCFcAEAAAAAAPhCuAAAAAAAAHwhXAAAAAAAAL4QLgAAAAAAAF8IFwAAAAAAgC+ECwAAAAAAwBfCBQAAAAAA4AvhAgAAAAAA8IVwAQAAAAAA+EK4AAAAAAAAfCFcAAAAAAAAvhAuAAAAAAAAXwgXAAAAAACAL4QLAAAAAADAF8IFAAAAAADgC+ECAAAAAADwhXABAAAAAAD4QrgAAAAAAAB8IVwAAAAAAAC+EC4AAAAAAABfCBcAAAAAAIAvhAsAAAAAAMAXwgUAAAAAAOAL4QIAAAAAAPCFcAEAAAAAAPhCuAAAAAAAAHwhXAAAAAAAAL4QLgAAAAAAAF8IFwAAAAAAgC+ECwAAAAAAwBfCBQAAAAAA4AvhAgAAAAAA8IVwAQAAAAAA+EK4AAAAAAAAfIlquGBmfczsWzP7zsxGlLN/iJl9Y2Z5ZjbPzNpHsx4AAAAAAFD1ohYumFmcpImSfimpvaRrygkPXnfOJTvnUiX9j6Qno1UPAAAAAACIjmjOXEiX9J1z7nvn3D5Jb0q6NLyBc2572MMTJbko1gMAAAAAAKKgehT7biJpbdjjdZK6lm5kZr+XdKekGpLOj2I9AAAAAAAgCo76go7OuYnOuVaS/iDpgfLamNlNZpZrZrkbN248sgUCAAAAAICDima4sF5Ss7DHTb1tFXlT0mXl7XDOPe+c6+yc69yoUaMqLBEAAAAAAPgVzXDhK0mnm1lLM6shqZ+k7PAGZnZ62MOLJa2KYj0AAAAAACAKorbmgnOu0MyGSvpQUpykl5xzS83sYUm5zrlsSUPNrJek/ZK2SBoQrXoAAAAAAEB0RHNBRznnZkqaWWrbyLDfb4vm+AAAAAAAIPqO+oKOAAAAAAAgthEuAAAAAAAAXwgXAAAAAACAL4QLAAAAAADAF8IFAAAAAADgC+ECAAAAAADwhXABAAAAAAD4QrgAAAAAAAB8IVwAAAAAAAC+EC4AAAAAAABfCBcAAAAAAIAvhAsAAAAAAMAXwgUAAAAAAOAL4QIAAAAAAPCFcAEAAAAAAPhCuAAAAAAAAHwhXAAAAAAAAL4QLgAAAAAAAF8IFwAAAAAAgC+ECwAAAAAAwBfCBQAAAAAA4AvhAgAAAAAA8IVwAQAAAAAA+EK4AAAAAAAAfCFcAAAAAAAAvhAuAAAAAAAAXwgXAAAAAACAL4QLAAAAAADAF8IFAAAAAADgC+ECAAAAAADwhXABAAAAAAD4QrgAAAAAAAB8IVwAAAAAAAC+EC4AAAAAAABfCBcAAAAAAIAvlYYLZnaVmdX1fn/AzN41s7TolwYAAAAAAGJBJDMX/uic22Fm50jqJelFSZOiWxYAAAAAAIgVkYQLB7w/L5b0vHPufUk1olcSAAAAAACIJZGEC+vN7DlJV0uaaWY1IzwOAAAAAAAcByIJCX4t6UNJmc65rZISJQ2PalUAAAAAACBmVBouOOd+kvSjpHO8TYWSVkWzKAAAAAAAEDsiuVvEg5L+IOleb1O8pNeiWRQAAAAAAIgdkVwWcbmkLEm7JMk5t0FS3WgWBQAAAAAAYkck4cI+55yT5CTJzE6MbkkAAAAAACCWRBIuvOXdLaK+md0oaZak/41uWQAAAAAAIFZUP9hOMzNJUyWdIWm7pLaSRjrnPj4CtQEAAAAAgBhw0HDBOefMbKZzLlkSgQIAAAAAACgjkssiFppZl6hXAgAAAAAAYtJBZy54ukq6zsx+UPCOEabgpIaUqFYGAAAAAABiQiThQmbUqwAAAAAAADGr0ssinHM/SKov6RLvp763DQAAAAAAoPJwwcxukzRF0snez2tmdmu0CwMAAAAAALEhkssifiupq3NulySZ2Z8l5UgaH83CAAAAAABAbIjkbhEm6UDY4wPeNgAAAAAAgIhmLrws6Usz+5v3+DJJL0avJAAAAAAAEEsqDRecc0+a2RxJ53ibBjnnvo5qVQAAAAAAIGZUGi6Y2VmSljrnFnqP65lZV+fcl1GvDgAAAAAAHPMiWXNhkqSdYY93etsAAAAAAAAiW9DROeeKHzjnihTZWg0AAAAAAOA4EEm48L2ZDTOzeO/nNknfR7swAAAAAAAQGyIJF4ZIypC0XtI6SV0l3RTNogAAAAAAQOyI5G4RP0rqdwRqAQAAAAAAMajSmQtm9j/eHSLizWy2mW00s98cieIAAAAAAMCxL5LLIno757ZL6ispX1JrScOjWRQAAAAAAIgdkYQLxZdOXCzpbefctijWAwAAAAAAYkwkt5ScYWYrJO2WdLOZNZK0J7plAQAAAACAWFHpzAXn3AgF7xbR2Tm3X9JPki6NdmEAAAAAACA2RDJzQc65zWG/75K0K2oVAQAAAACAmBLJmgsAAAAAAAAVIlwAAAAAAAC+VBoumNnlZpYQ9ri+mV0W3bIAAAAAAECsiGTmwoPht590zm2V9GD0SgIAAAAAALEkknChvDYRLQQJAAAAAAB+/iIJF3LN7Ekza+X9PClpQbQLAwAAAAAAsSGScOFWSfskTfV+9kr6fTSLAgAAAAAAsaPSyxucc7skjTgCtQAAAAAAgBhUYbhgZk875243s/+T5Ervd85lRbUyAAAAAAAQEw42c+FV78+xR6IQAAAAAAAQmyoMF5xzC8wsTtJNzrnrjmBNAAAAAAAghhx0QUfn3AFJzc2sxhGqBwAAAAAAxJhKF3SU9L2kz80sW9Ku4o3OuSejVhUAAAAAAIgZkYQLq72fapLqetvKLPAIAAAAAACOT5GEC8ucc2+HbzCzq6JUDwAAAAAAiDEHXXPBc2+E2wAAAAAAwHGowpkLZvZLSRdJamJm48J21ZNUGO3CAAAAAABAbDjYZREbJOVKypK0IGz7Dkl3RLMoAAAAAAAQOyoMF5xziyQtMrPXvXa/cM59e8QqAwAAAAAAMSGSNRf6SMqT9HdJMrNU77aUAAAAAAAAEYULoySlS9oqSc65PEkto1gTAAAAAACIIZGEC/udc9tKbXPRKAYAAAAAAMSegy3oWGypmV0rKc7MTpc0TFIgumUBAAAAAIBYEcnMhVslnSlpr6Q3JG2XdHs0iwIAAAAAALGj0pkLzrmfJN3v/QAAAAAAAJRQabhgZp0l3SepRXh751xK9MoCAAAAAACxIpI1F6ZIGi7pG0lF0S0HAAAAAADEmkjChY3OueyoVwIAAAAAAGJSJOHCg2b2gqTZCi7qKElyzr0btaoAAAAAAEDMiCRcGCTpDEnx+u9lEU4S4QIAAAAAAIgoXOjinGsb9UoAAAAAAEBMqhZBm4CZtY96JQAAAAAAICZFMnPhLEl5ZvZPBddcMEmOW1ECAAAAAAApsnChT9SrAAAAAAAAMavScME598ORKAQAAAAAAMSmSNZcAAAAAAAAqBDhAgAAAAAA8IVwAQAAAAAA+EK4AAAAAAAAfCFcAAAAAAAAvhAuAAAAAAAAXwgXAAAAAACAL4QLAAAAAADAF8IFAAAAAADgC+ECAAAAAADwJarhgpn1MbNvzew7MxtRzv47zWyZmS02s9lm1jya9QAAAAAAgKoXtXDBzOIkTZT0S0ntJV1jZu1LNftaUmfnXIqkaZL+J1r1AAAAAACA6IjmzIV0Sd855753zu2T9KakS8MbOOc+cc795D38QlLTKNYDAAAAAACiIJrhQhNJa8Mer/O2VeS3kj4ob4eZ3WRmuWaWu3HjxiosEQAAAAAA+HVMLOhoZr+R1FnSmPL2O+eed851ds51btSo0ZEtDgAAAAAAHFT1KPa9XlKzsMdNvW0lmFkvSfdL6uGc2xvFegAAAAAAQBREc+bCV5JON7OWZlZDUj9J2eENzKyjpOckZTnnfoxiLQAAAAAAIEqiFi445wolDZX0oaTlkt5yzi01s4fNLMtrNkZSHUlvm1memWVX0B0AAAAAADhGRfOyCDnnZkqaWWrbyLDfe0VzfAAAAAAAEH3HxIKOAAAAAAAgdhEuAAAAAMD/b+/eo+yq6nzRfyfESPMSAsbrqQLSscirMMSYiKCIoJgYpRANEODQYPAi3TgQD9jobYnIOAyhr2C3Nz6OigeaASmebfAorw6HC/ZQkwBBIA3hES4kKmAM2OEVkqz7R1XKKpJAklW7HuHzGSODvR571W9qrdp7fdeccwG1CBcAAACAWoQLAAAAQC3CBQAAAKAW4QIAAABQi3ABAAAAqEW4AAAAANQiXAAAAABqES4AAAAAtQgXAAAAgFqECwAAAEAtwgUAAACgFuECAAAAUItwAQAAAKhFuAAAAADUIlwAAAAAahEuAAAAALUIFwAAAIBahAsAAABALcIFAAAAoBbhAgAAAFCLcAEAAACoRbgAAAAA1CJcAAAAAGoRLgAAAAC1CBcAAACAWoQLAAAAQC3CBQAAAKAW4QIAAABQi3ABAAAAqEW4AAAAANQiXAAAAABqES4AAAAAtQgXAAAAgFqECwAAAEAtwgUAAACgFuECAAAAUItwAQAAAKhFuAAAAADUIlwAAAAAahEuAAAAALUIFwAAAIBahAsAAABALcIFAAAAoBbhAgAAAFCLcAEAAN4Ebr755owePTotLS258MILN9h+5513ZuLEiRkyZEiuu+66Dbb/+c9/TnNzc77whS8kSf7zP/8zEyZM6Pq355575swzz2x4O4CBSbgAAMCg0dsXyEmyevXqnHrqqRk1alTGjBmT66+/vqFt6A9r167N6aefnptuuimLFy/OnDlzsnjx4h777L333rnsssty/PHHb/QY5557bj70oQ91Le+yyy5ZtGhR17999tknn/70pxvaDmDgEi4AADAoNOICOUkuuOCCDB8+PEuWLMnixYtzyCGHNKwN/WX+/PlpaWnJyJEjM3To0MyYMSNz587tsc+IESMyfvz4bLfdhpcId999d55++ul87GMf2+jxlyxZkmeeeSYHH3xwQ+oHBj7hAgAAg0KjLpB/8pOf5Ktf/WqSZLvttsuee+7ZuEb0k+XLl2evvfbqWm5ubs7y5cs3673r1q3LWWedlW9961ub3Ke9vT3HHntsSim1awUGJ+ECAACDQiMukJ977rkkHT0aJk6cmKOPPjpPP/107xW9Dfje976XadOmpbm5eZP7tLe357jjjuvDqoCBRrgAAMA2b1MXyGvWrMmyZcty0EEH5Z577smBBx6Ys88+u5+qbJympqY89dRTXcvLli1LU1PTZr33V7/6VWbPnp0RI0bk7LPPzr/8y7/kK1/5Stf2++67L2vWrMl73/veXq8bGqkRc7hMnTo1+++/f1pbW3Paaadl7dq1DW3DQDKkvwsAAIDNUfcC+a677sr3vve9rFq1KqtXr87OO++cb37zm9lxxx27JiI8+uijc+mllzak/v40efLkPPLII1m6dGmamprS3t6eq666arPee+WVV3a9vuyyy7Jw4cIeF2Jz5szRa4FBZ/0cLrfddluam5szefLktLW1Zdy4cV37rJ/DZVNDgjY2h8s111yTXXfdNVVVZfr06bn22mszY8aMhrZloNBzAQCAQaH7BfLq1avT3t6etra2zXrvlVdemSeffDJPPPFEvvWtb+Vv/uZvcuGFF6aUkiOOOCJ33HFHkmTevHk9Li62FUOGDMns2bMzZcqUjB07Nsccc0xaW1sza9as3HjjjUmSBQsWpLm5Oddee20+//nPp7W1dbOOfc011wgXGHQaNYfLrrvumqSjV9Tq1avfVPOQCBcAABgUGnWBfNFFF+W8887L+PHjc8UVV+Tiiy9udFP6xbRp07JkyZI89thj+Yd/+Ickyfnnn98V0EyePDnLli3LCy+8kBUrVuTBBx/c4Bgnn3xyZs+e3WPd448/njFjxjS+AbyhRnTz31Y1cpLTKVOmZPjw4dlll10yffr0Xql3MDAsAgCAQWPatGmZNm1aj3Xnn39+1+v1F8iv5+STT87JJ5/ctbzPPvvkzjvv7NU6oa81qps/G3qjSU5vueWWvPzyyznhhBNy++235/DDD+/jCvuHcAEAAGCQ697NP0lXN//u4cKIESOS5HW7+U+dOjULFy7sk5r7UyPmcOneW2SHHXbIkUcemblz575pwgXDIgAAAAa5Rnbz3xY1Yg6XVatW5fe//32SjjkXfv7zn7+phgzpuQAAANuIfUeN7u8SNumRJQ/3dwlswht1898WdZ/DZe3atZk5c2bXHC6TJk1KW1tbFixYkKOOOiorV67Mz372s3z961/f6Fwk673wwgtpa2vLK6+8knXr1uXQQw/Naaed1oet6l/CBQAABpZbB/Ds6h+r+rsC2KhGd/PfFvX2HC7veMc7smDBgl6vc7AQLgAAAAxy3bv5NzU1pb29PVddddVmvffKK6/sen3ZZZdl4cKF23ywQO8z5wIAAMAg16hHtcLmKlU1uLp2TZo0qXozzF7aly6+bs/+LmGTzpr+x/4uAQDoa4ZFbDVzLgCNVEq5u6qqSRvbZlgEAAAA25yBHLYl217gZlgEAAAAUItwAQAAAKjFsAgAAICBzDwkDAJ6LgAAAAC1CBcAAACAWoQLAAAAQC3CBQAAAKAW4QIAAABQi3ABAAAAqEW4AAAAANQiXAAAAABqES4AAAAAtQgXAAAAgFqECwAAAEAtwgUAAACgFuECAAAAUItwAQAAAKhFuAAAAADUIlwAAAAAahEuAAAAALUIFwAAAIBahAsAAABALcIFAAAAoBbhAgAAAFCLcAEYUG6++eaMHj06LS0tufDCCzfYfuedd2bixIkZMmRIrrvuuh7bpk6dmt122y2f/OQne6w/4YQTMnr06Oy3336ZOXNmXn311Ya2AQAA3myECwxajbgInT17dlpaWlJKyR//+MeG1s+G1q5dm9NPPz033XRTFi9enDlz5mTx4sU99tl7771z2WWX5fjjj9/g/V/+8pdzxRVXbLD+hBNOyEMPPZT7778/L730Un784x83rA0AAPBmJFxgUGrURegHPvCB/Nu//Vv22WefhtXOps2fPz8tLS0ZOXJkhg4dmhkzZmTu3Lk99hkxYkTGjx+f7bbb8M/XRz7ykeyyyy4brJ82bVpKKSml5H3ve1+WLVvWsDYAAMCbkXBhgGnE3fhtUaMuQt/znvdkxIgRjSqbN7B8+fLstddeXcvNzc1Zvnx5rx3/1VdfzRVXXJGpU6f22jEBAADhwoDSqLvx26JGX4Sybfq7v/u7fOhDH8rBBx/c36U0RJ1w8vLLL8++++6bfffdd6kugQAAG0xJREFUN5dffnnX+quvvjrjx49Pa2trzjnnnIa3AQCAwUm4MIA06m48DBZNTU156qmnupaXLVuWpqamXjn2N77xjTz77LO55JJLeuV4A02dcPJPf/pTvvGNb+Q3v/lN5s+fn2984xtZuXJlVqxYkS9/+cuZN29eHnzwwfzhD3/IvHnz+rJZAAAMEsKFAcTd+M3XyItQ+s/kyZPzyCOPZOnSpVm9enXa29vT1tZW+7g//vGPc8stt2TOnDkbDea2BXXCyVtuuSWHH354hg0blt133z2HH354br755jz++OPZd9998/a3vz1J8tGPfjTXX399n7UJAIDBY9v8ls02r1EXofSvIUOGZPbs2ZkyZUrGjh2bY445Jq2trZk1a1ZuvPHGJMmCBQvS3Nyca6+9Np///OfT2tra9f6DDz44Rx99dObNm5fm5ubccsstSZLTTjstTz/9dA488MBMmDAh559/fr+0r5HqhJObem9LS0sefvjhPPHEE1mzZk1++tOf9gj1AABgvSH9XQB/4W785ut+Ebp27drMnDmz6yJ00qRJaWtry4IFC3LUUUdl5cqV+dnPfpavf/3refDBB5N0XIQ+9NBDWbVqVZqbm3PppZdmypQp+c53vpN//Md/zB/+8IeMHz8+06ZN89jCPjZt2rRMmzatx7ruYcDkyZM3+bSHu+66a6Pr16xZ03sFvonsvvvu+f73v59jjz022223XQ466KA89thj/V0WAAADkHBhAOl+N76pqSnt7e256qqr+rusAasRF6FnnHFGzjjjjN4rEvpInXCyqakpd9xxR4/3fvjDH06SHHHEETniiCOSJD/84Q+z/fbb91rNAABsOwyLGEAa1SUc2PbVGSo0ZcqU3HrrrVm5cmVWrlyZW2+9NVOmTEmSPPPMM0mSlStX5nvf+14+97nPNawNAAAMXnouDDCNuBsPbPvqDBUaNmxYzj333EyePDlJMmvWrAwbNixJ8sUvfjH33Xdf1/pRo0b1WxsBABi4SlVV/V3DFpk0aVK1cOHC/i5jm3LxdXv2dwmbdNb0P/Z3CTTCraW/K3h9HxtcfxcBtjkD+XNigH9G7DtqdH+XsEmPLHm4v0sYvJwTW2Ugnw/J4DwnSil3V1U1aWPb9FxgQBvIfxAG4x8DAACARjDnAgAAAFCLngt9YSB3Y0qS7NHfBQAAADCICRcABhFDhQAAGIgMiwAAAABqES4AAAAAtQgXAAAAgFqECwAAAEAtwgUAAACgFuECALBNuvnmmzN69Oi0tLTkwgsv3GD7K6+8kmOPPTYtLS054IAD8sQTTyRJVq9enc9+9rN597vfnf333z933HHHBu9ta2vLfvvt1+AWAMDgIVwAALY5a9euzemnn56bbropixcvzpw5c7J48eIe+1x66aXZfffd8+ijj+ZLX/pSzjnnnCTJj370oyTJ/fffn9tuuy1nnXVW1q1b1/W+G264ITvvvHPfNQYABoGGhgullKmllIdLKY+WUr6yke0fKqXcU0pZU0qZ3shaAGAw6u277y+++GI+8YlPZMyYMWltbc1XvrLBx/M2Yf78+WlpacnIkSMzdOjQzJgxI3Pnzu2xz9y5c3PSSSclSaZPn5558+alqqosXrw4hx12WJJk+PDh2W233bJw4cIkyapVq3LJJZfka1/7Wt82CAAGuIaFC6WU7ZN8N8nHk4xLclwpZdxrdnsyyclJrmpUHQAwWDXq7vvZZ5+dhx56KPfee2/+/d//PTfddFPfNqwPLF++PHvttVfXcnNzc5YvX77JfYYMGZK3ve1tWbFiRfbff//ceOONWbNmTZYuXZq77747Tz31VJLk3HPPzVlnnZUdd9yx7xoDAINAI3suvC/Jo1VVPV5V1eok7UmO7L5DVVVPVFX12yTrNnYAAHgza8Td9x133DGHHnpokmTo0KGZOHFili1b1rcNG+BmzpyZ5ubmTJo0KWeeeWYOOuigbL/99lm0aFEee+yxHHXUUf1dIgAMOI0MF5qSPNVteVnnui1WSjm1lLKwlLLw2Wef7ZXiAGCga9Td9/Wee+65/OxnP8tHPvKRxjemjzU1NfVo77Jly9LU1LTJfdasWZPnn38+e+yxR4YMGZJvf/vbWbRoUebOnZvnnnsuo0aNyq9+9assXLgwI0aMyAc/+MEsWbIkH/7wh/uyWQAwYA2KCR2rqvphVVWTqqqa9Pa3v72/ywGAAW9Td9/XW7NmTY477ricccYZGTly5Fb/nK2dE+LVV1/NSSedlHe/+90ZO3ZsvvnNb3a955//+Z+z3377pbW1Nf/0T/+0VXVNnjw5jzzySJYuXZrVq1envb09bW1tPfZpa2vL5ZdfniS57rrrcthhh6WUkhdffDEvvPBCkuS2227LkCFDMm7cuPzt3/5tfve73+WJJ57IL3/5y4waNWqjT5IAgDejIQ089vIke3Vbbu5cBwBshi25+97c3Nzj7nspJd/+9re79jvooIMyatSoruVTTz01++67b84888ytrm/9nBC33XZbmpubM3ny5LS1tWXcuL9MsdR9Toj29vacc845ufrqq3PttdfmlVdeyf33358XX3wx48aNy3HHHZdVq1blRz/6UebPn5+hQ4dm6tSp+eQnP5mWlpYtqm3IkCGZPXt2pkyZkrVr12bmzJlpbW3NrFmzMmnSpLS1teWUU07JiSeemJaWlgwbNizt7e1JkmeeeSZTpkzJdtttl6amplxxxRVb/b8RALxZNDJcWJBk31LKX6cjVJiR5PgG/jwA2KZ0v/ve1NSU9vb2XHVVzzmQ1999P/DAAze4+15VVXbaaaced9+T5Gtf+1qef/75/PjHP65VX/c5IZJ0zQnRPVyYO3duzjvvvCQdc0J84QtfSFVVKaXkhRdeyJo1a/LSSy9l6NCh2XXXXbNgwYIccMABXRMmHnLIIbnhhhvy93//91tc37Rp0zJt2rQe684///yu1zvssEOuvfbaDd43YsSIPPzww6977BEjRuSBBx7Y4poAYFvVsGERVVWtSfKFJLck+Y8k11RV9WAp5fxSSluSlFIml1KWJTk6yf8opTzYqHoAYLDpfvd97NixOeaYY7ruvt94441JklNOOSUrVqxIS0tLLrnkkq6hCc8880wmTpyYsWPH5qKLLuq6+75s2bJccMEFWbx4cSZOnJgJEyZsdchQZ06I6dOnZ6eddso73/nO7L333jn77LMzbNiw7LfffrnrrruyYsWKvPjii/nFL36xwVwRAMDA08ieC6mq6hdJfvGadbO6vV6QjuESAMBG9Pbd9+bm5lRV1fuFbqH58+dn++23z+9+97usXLkyBx98cD760Y9m7NixOeecc/Kxj30sO+20UyZMmNBjrggAYGAaFBM6AgADT50nMlx11VWZOnVq3vKWt2T48OH5wAc+kIULFybp6I1x9913584778zuu+/eY64IAGBgamjPBQBg21VnToi99947t99+e0488cS88MIL+fWvf901ueQzzzyT4cOH58knn8wNN9yQX//615usYd9RoxvaxjoeWfL68zYAwLZEzwUAYKvUmRPi9NNPz6pVq9La2prJkyfns5/9bMaPH58k+cxnPpNx48bliCOOyHe/+93stttu/dZG2BK9/WjWl19+Oe973/uy//77p7W1NV//+tf7sjkAW0TPBQDob7eW/q5g0z72+vMzbO2cEDvvvPNG1yfJXXfdtRWFQv9qxKNZ99lnn9x+++3Zeeed8+qrr+aDH/xgPv7xj+f9739/P7YUYOP0XAAAgJq6P5p16NChXY9m7W7u3Lk56aSTknQ8mnXevHmv+2jWUkp23nnnJB29G1599dWUMoDDSOBNTbgAAAA1NeLRrElHj4gJEyZk+PDhOfzww3PAAQf0XaMAtoBwAQAA+lH3R7MuXbo0F198cR5//PEkyfbbb59FixZl2bJlmT9/fh544IF+rhZg48y5AABs0sXX7dnfJbyBPfq7AEiyZY9mbW5u3qxHs44cObLrvbvttlsOPfTQ3Hzzzdlvv/36rF0Am0vPBQAAqKn7o1lXr16d9vb2tLW19dhn/aNZk2z00axJuh7NOmbMmDz77LN57rnnkiQvvfRSbrvttowZM6ZvGwawmfRcAACAmro/mnXt2rWZOXNm16NZJ02alLa2tpxyyik58cQT09LSkmHDhqW9vT1Jx6NZP/vZz6a1tTVVVXU9mvW3v/1tTjrppKxduzbr1q3LMccck09+8pP93FKAjRMuAABAL+jtR7OOHz8+9957b+8XCtAAhkUAAAAAtQgXAAAAgFqECwAAAEAt5lwAAIDN5PGsABun5wIAAABQi3ABAAAAqEW4AAAAANQiXAAAAABqES4AAAAAtQgXAAAAgFqECwAAAEAtwgUAAACgFuECAAAAUItwAQAAAKhFuAAAAADUIlwAAAAAahEuAAAAALUIFwAAAIBahAsAAABALcIFAAAAoBbhAgAAAFCLcAEAAACoRbgAAAAA1CJcAAAAAGoRLgAAANDrbr755owePTotLS258MILN9j+yiuv5Nhjj01LS0sOOOCAPPHEE0mSJ554In/1V3+VCRMmZMKECTnttNO63rN69eqceuqpGTVqVMaMGZPrr7++r5rDGxjS3wUAAACwbVm7dm1OP/303HbbbWlubs7kyZPT1taWcePGde1z6aWXZvfdd8+jjz6a9vb2nHPOObn66quTJO9617uyaNGiDY57wQUXZPjw4VmyZEnWrVuXP/3pT33WJl6fngsAAAD0qvnz56elpSUjR47M0KFDM2PGjMydO7fHPnPnzs1JJ52UJJk+fXrmzZuXqqpe97g/+clP8tWvfjVJst1222XPPfdsTAPYYsIFAAAAetXy5cuz1157dS03Nzdn+fLlm9xnyJAhedvb3pYVK1YkSZYuXZr3vOc9OeSQQ3LXXXclSZ577rkkybnnnpuJEyfm6KOPztNPP90XzWEzCBcAtkAjxg5OnTo1+++/f1pbW3Paaadl7dq1fdUcAIAB553vfGeefPLJ3Hvvvbnkkkty/PHH589//nPWrFmTZcuW5aCDDso999yTAw88MGeffXZ/l0sn4QLAZlo/dvCmm27K4sWLM2fOnCxevLjHPt3HDn7pS1/KOeec07Vt/djBRYsW5Qc/+EHX+muuuSb33XdfHnjggTz77LO59tpr+6xNAACN0NTUlKeeeqpredmyZWlqatrkPmvWrMnzzz+fPfbYI29961uzxx57JEne+9735l3veleWLFmSPfbYIzvuuGM+/elPJ0mOPvro3HPPPX3UIt6IcAFgMzVq7OCuu+6apONDdfXq1SmlNKYBAAB9ZPLkyXnkkUeydOnSrF69Ou3t7Wlra+uxT1tbWy6//PIkyXXXXZfDDjsspZQ8++yzXT05H3/88TzyyCMZOXJkSik54ogjcscddyRJ5s2b12OCSPqXcAFgMzVi7OB6U6ZMyfDhw7PLLrtk+vTpDW4JAEBjDRkyJLNnz86UKVMyduzYHHPMMWltbc2sWbNy4403JklOOeWUrFixIi0tLbnkkku6hpzeeeedGT9+fCZMmJDp06fnBz/4QYYNG5Ykueiii3Leeedl/PjxueKKK3LxxRf3WxvpyaMoAfrA+rGDe+yxR+6+++586lOfyoMPPtjVa+GWW27Jyy+/nBNOOCG33357Dj/88H6uGACgnmnTpmXatGk91p1//vldr3fYYYeNDgf9zGc+k8985jMbPeY+++yTO++8s3cLpVfouQCwmRoxdrC7HXbYIUceeeQGQy0AAGCgEy4AbKZGjB1ctWpVfv/73yfpCCN+/vOfZ8yYMX3bMAAAqMmwCIDN1H3s4Nq1azNz5syusYOTJk1KW1tbTjnllJx44olpaWnJsGHD0t7enqRj7OCsWbPylre8Jdttt13X2MGnn346bW1teeWVV7Ju3boceuihPR5TCQAAg4FwAWAL9PbYwXe84x1ZsGBB7xcKANAHLr5uz/4u4XXs0d8FvKkYFgEAAADUIlwAAAAAahEuAAAAALWYcwHgNYwdBACALaPnAgAAAFCLcAEAAACoRbgAAAAA1CJcAAAAAGoRLgAAAAC1CBcAAACAWoQLAAAAQC3CBQAAAKAW4QIAAABQi3ABAAAAqEW4AAAAANQiXAAAAABqES4AAAAAtQgXAAAAgFqECwAAAEAtwgUAAACgFuECAAAAUItwAQAAAKhFuAAAAADUIlwAAAAAahEuAAAAALUIFwAAAIBahAsAAABALcIFAAAAoBbhAgAAAFCLcAEAAACoRbgAAAAA1CJcAAAAAGoRLgAAAAC1CBcAAACAWoQLAAAAQC3CBQAAAKAW4QIAAABQi3ABAAAAqEW4AAAAANQiXAAAAABqES4AAAAAtQgXAAAAgFqECwAAAEAtwgUAAACgFuECAAAAUItwAQAAAKhFuAAAAADUIlwAAAAAahEuAAAAALUIFwAAAIBahAsAAABALcIFAAAAoBbhAgAAAFCLcAEAAACoRbgAAAAA1CJcAAAAAGoRLgAAAAC1CBcAAACAWoQLAAAAQC3CBQAAAKAW4QIAAABQi3ABAAAAqEW4AAAAANQiXAAAAABqaWi4UEqZWkp5uJTyaCnlKxvZ/tZSytWd239TShnRyHoAAACA3tewcKGUsn2S7yb5eJJxSY4rpYx7zW6nJFlZVVVLkm8nuahR9QAAAACN0cieC+9L8mhVVY9XVbU6SXuSI1+zz5FJLu98fV2Sj5RSSgNrAgAAAHrZkAYeuynJU92WlyU5YFP7VFW1ppTyfJI9kvyx+06llFOTnNq5uKqU8nBDKn7TWtGbB9szr/n/r55era1XycG2Zc6JreGc2Fb1+u+cc4JBzjmxtZwT26qBfE4M3PMhGbTnxD6b2tDIcKHXVFX1wyQ/7O86eGOllIVVVU3q7zpgoHBOQE/OCejJOQE9OScGr0YOi1ieZK9uy82d6za6TyllSJK3ZaDHSwAAAEAPjQwXFiTZt5Ty16WUoUlmJLnxNfvcmOSkztfTk9xeVVXVwJoAAACAXtawYRGdcyh8IcktSbZP8pOqqh4spZyfZGFVVTcmuTTJFaWUR5P8KR0BBIOb4SvQk3MCenJOQE/OCejJOTFIFR0FAAAAgDoaOSwCAAAAeBMQLgAAAAC1CBfYKqWUT5VSqlLKmM7lEaWUl0op95ZS/qOUMr+UcnK3/U8upczut4KhF5RS/o9SSnsp5bFSyt2llF+UUkZ1bjuzlPJyKeVtr3nPx0spC0spizvPj4s715/XeQ61dNv3zM51Hr/EoLWp86SU8kDn9h1LKVeWUu4vpTxQSvllKWXnzm2r+rd6eH2v9zlQ87hvL6X8pvNz4uDO4+72Bu95opSy50bWn1dKObtuTdDbNvY3vvvvaynl/Z3nwaLO64nzXrPvP5VSlpdSXMMOUA2b0JFt3nFJftn53693rnusqqr3JEkpZWSSG0oppaqq/9lPNUKvKaWUJP+a5PKqqmZ0rts/yTuSLEnHubAgyaeT/M/O7fslmZ3kE1VVPVRK2T7Jqd0Oe386JrL9753LRyd5sPGtgcZ4g/NkvS8mebqqqnd3bh+d5NW+rhW21GZ8DqSUMqSqqjVbcfiPJLm/qqrPdS7f1Qslw2BzeZJjqqq6r/M70+j1GzoDhaOSPJXkkCT/u39K5PVIfdhinXeYPpjklGziCR9VVT2e5L8lOaMPS4NGOjTJq1VV/WD9iqqq7quq6q5SyruS7Jzka+kIGdb7+yQXVFX1UOf+a6uq+n637T9NcmSSdB7j+SR/bGwzoKE2ep6k48vgeu9Msrzb9oerqnql70qErbap3+/tSyl3lVJuTLI4SUopP+3s2fBgKaUrVC6lrCqlXFBKua+U8utSyjtKKROS/GOSIzvv2P5V914JpZT/2tkjdFEp5X90XnT1UEr5h1LKklLKL9PtggwGmeFJfp90fWda3G3bh9NxA+b76fldiwFEuMDWODLJzVVVLUmyopTy3k3sd0+SMX1XFjTUfknu3sS2GUna03GnaXQp5R2b8Z4k+XOSpzp7OMxIcnUv1Qr95Y1+55PkJ0nOKaX8qpTy30sp+/ZBXdAbXu/3e2KSL1ZVtX6IxMyqqt6bZFKSM0ope3Su3ynJr6uq2j/JnUn+z6qqFiWZleTqqqomVFX10vqDllLGJjk2yQeqqpqQZG2SE7r/4M7vYTOSTEgyLcnk+k2FfvHtJA+XUv61lPL5UsoO3bYdl2ROOnoPfaKU8pZ+qZDXJVxgaxyXjgupdP53U+lh6ZtyoN8dl6S9qqp1Sa5Px/CGzdWeji+Fn0rHByZs0zovpEYm+b+TDEuyoPMCCgaz+VVVLe22fEYp5b4kv06yV5L1IdrqJP+r8/XdSUa8wXE/kuS96ThPFnUuj3zNPgcn+deqql6squrPSW7c6lZAP6qq6vx0BHK3Jjk+yc1JUkoZmo7g7Kedv+O/STKlv+pk08y5wBYppQxLcliSd5dSqiTbJ6mSfHcju78nyX/0YXnQSA8mmf7alaWUd6fjS+NtHcNxMzTJ0nTMtfBgOr4U3vc6x/1f6bjIWlhV1Z87jwGD1UbPk9eqqmpVkhvSMTfPunR8afR5wUD3er/fL6x/UUr5cJKPJjmwqqoXSyl3JFl/B/bVqqqqztdr88bfxUs65nj46tYWDYNJVVWPJfl+KeVHSZ7t7PVzUJLdktzf+T1pxyQv5S9BHQOEngtsqelJrqiqap+qqkZUVbVXOi6k9uq+UyllRJJvJfl/+rxCaIzbk7z1NWNnxyf5TpLzOs+HEVVV/Zck/6WUsk86QoP/q/zliRLblVJO637QqqpeTHJOkgv6qiHQQJs6T/bqtvyBUsruna+HJhmX5P/r60JhK2zq9/vg1+z3tiQrO4OFMUneX+NnzksyvZQyvPPnDev8fOnuziSf6pyrYZckR9T4edBvSimfKH+5y7JvOgK459LRQ/Rz679rJfnrJIeXUnbsn0rZFOECW+q4bNh1+/okX03yrs5HKP1HkmuSfMeTIthWdN5pOirJRzsfQfZgkm+mY4Kh154T/5pkRlVVv01yZpI5nefFA9mwO2uqqmqvquqeRtYPfeF1zpM/dNvtXUn+31LK/UnuTbIwHZ8jMKBt5u930tGVe0jn3/0L0zE0Ymt/5uJ0TBZ8aynlt0luS8ekqN33uScdc/bcl+SmdDy5CAaiHUspy7r9+2+v2X5iOuZcWJTkinTML/LWJFOT/Hz9TlVVvZCOp9YJ0gaY8peeWQAAAABbTs8FAAAAoBbhAgAAAFCLcAEAAACoRbgAAAAA1CJcAAAAAGoRLgAAAAC1CBcAAACAWv5/GeL/4nXaqIEAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 1152x720 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SsfU8fpQ1qRs"
      },
      "source": [
        "**Precision on the 'text' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 531
        },
        "id": "BzrUyXLR1qRs",
        "outputId": "9b7f8cfa-648b-4823-e3b0-ff0559e2896b"
      },
      "source": [
        "import matplotlib\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "match_metrics_precision_text = []\n",
        "match_metrics_precision_text.append(round(adi_res_match_precision_text['metric_score'], 3))\n",
        "match_metrics_precision_text.append(round(cacm_res_match_precision_text['metric_score'], 3))\n",
        "match_metrics_precision_text.append(round(cisi_res_match_precision_text['metric_score'], 3))\n",
        "match_metrics_precision_text.append(round(cran_res_match_precision_text['metric_score'], 3))\n",
        "match_metrics_precision_text.append(round(lisa_res_match_precision_text['metric_score'], 3))\n",
        "match_metrics_precision_text.append(round(med_res_match_precision_text['metric_score'], 3))\n",
        "match_metrics_precision_text.append(round(npl_res_match_precision_text['metric_score'], 3))\n",
        "match_metrics_precision_text.append(round(time_res_match_precision_text['metric_score'], 3))\n",
        "\n",
        "bert_text_metrics_precision = []\n",
        "bert_text_metrics_precision.append(round(adi_res_bert_text_precision['metric_score'], 3))\n",
        "bert_text_metrics_precision.append(round(cacm_res_bert_text_precision['metric_score'], 3))\n",
        "bert_text_metrics_precision.append(round(cisi_res_bert_text_precision['metric_score'], 3))\n",
        "bert_text_metrics_precision.append(round(cran_res_bert_text_precision['metric_score'], 3))\n",
        "bert_text_metrics_precision.append(round(lisa_res_bert_text_precision['metric_score'], 3))\n",
        "bert_text_metrics_precision.append(round(med_res_bert_text_precision['metric_score'], 3))\n",
        "bert_text_metrics_precision.append(round(npl_res_bert_text_precision['metric_score'], 3))\n",
        "bert_text_metrics_precision.append(round(time_res_bert_text_precision['metric_score'], 3))\n",
        "\n",
        "rerank_bert_text_metrics_precision = []\n",
        "rerank_bert_text_metrics_precision.append(round(adi_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(cacm_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(cisi_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(cran_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(lisa_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(med_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(npl_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(time_res_rerank_text_precision['metric_score'], 3))\n",
        "\n",
        "\n",
        "labels = ['ADI','CACM', 'CISI','Cranfield', 'LISA', 'Medline', 'NPL', 'Time']\n",
        "\n",
        "x = np.arange(len(labels))*2.5  # the label locations\n",
        "\n",
        "width = 0.6  # the width of the bars\n",
        "\n",
        "# orange, green, turquoise, black, red, yellow, white\n",
        "pragma_colors =  ['#ffb900', '#8cab13', '#22ab82', '#242526', '#cc0000',  '#ffcc00', '#ffffff']\n",
        "\n",
        "fig, ax = plt.subplots()\n",
        "rects1 = ax.bar(x - width, match_metrics_precision_text , width, label='With multi-match query on text', color=pragma_colors[0])\n",
        "rects2 = ax.bar(x, bert_text_metrics_precision, width, label='With BERT on text field', color=pragma_colors[1])\n",
        "rects3 = ax.bar(x + width, rerank_bert_text_metrics_precision, width, label='With re-ranking on text field', color=pragma_colors[3])\n",
        "\n",
        "# Add some text for labels, title and custom x-axis tick labels, etc.\n",
        "ax.set_ylabel('metric scores')\n",
        "ax.set_title('Precision scores on text field by corpus')\n",
        "ax.set_xticks(x)\n",
        "ax.set_xticklabels(labels)\n",
        "ax.legend()\n",
        "\n",
        "def autolabel(rects):\n",
        "    \"\"\"Attach a text label above each bar in *rects*, displaying its height.\"\"\"\n",
        "    for rect in rects:\n",
        "        height = rect.get_height()\n",
        "        ax.annotate('{}'.format(height),\n",
        "                    xy=(rect.get_x() + rect.get_width() / 2, height),\n",
        "                    xytext=(0, 3),  # 3 points vertical offset\n",
        "                    textcoords=\"offset points\",\n",
        "                    ha='center', va='bottom')\n",
        "\n",
        "autolabel(rects1)\n",
        "autolabel(rects2)\n",
        "autolabel(rects3)\n",
        "\n",
        "fig.tight_layout()\n",
        "fig.set_figwidth(16)\n",
        "fig.set_figheight(8)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x576 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "p87VCP7h1qRt"
      },
      "source": [
        "**Precision on a complex re-ranking query**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 647
        },
        "id": "vqca_N5-1qRt",
        "outputId": "3c022d92-f3a3-4c5e-8415-3f8e2c2aa4f1"
      },
      "source": [
        "import matplotlib\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "match_metrics_precision_text_title = []\n",
        "match_metrics_precision_text_title.append(round(adi_res_match_precision_text_title['metric_score'], 3))\n",
        "match_metrics_precision_text_title.append(round(cacm_res_match_precision_text_title['metric_score'], 3))\n",
        "match_metrics_precision_text_title.append(round(cisi_res_match_precision_text_title['metric_score'], 3))\n",
        "match_metrics_precision_text_title.append(round(cran_res_match_precision_text_title['metric_score'], 3))\n",
        "match_metrics_precision_text_title.append(round(lisa_res_match_precision_text_title['metric_score'], 3))\n",
        "\n",
        "complex_rerank_bert_metrics_precision = []\n",
        "complex_rerank_bert_metrics_precision.append(round(adi_res_complex_rerank_precision['metric_score'], 3))\n",
        "complex_rerank_bert_metrics_precision.append(round(cacm_res_complex_rerank_precision['metric_score'], 3))\n",
        "complex_rerank_bert_metrics_precision.append(round(cisi_res_complex_rerank_precision['metric_score'], 3))\n",
        "complex_rerank_bert_metrics_precision.append(round(cran_res_complex_rerank_precision['metric_score'], 3))\n",
        "complex_rerank_bert_metrics_precision.append(round(lisa_res_complex_rerank_precision['metric_score'], 3))\n",
        "\n",
        "rerank_bert_text_metrics_precision = []\n",
        "rerank_bert_text_metrics_precision.append(round(adi_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(cacm_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(cisi_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(cran_res_rerank_text_precision['metric_score'], 3))\n",
        "rerank_bert_text_metrics_precision.append(round(lisa_res_rerank_text_precision['metric_score'], 3))\n",
        "\n",
        "rerank_bert_title_metrics_precision = []\n",
        "rerank_bert_title_metrics_precision.append(round(adi_res_rerank_title_precision['metric_score'], 3))\n",
        "rerank_bert_title_metrics_precision.append(round(cacm_res_rerank_title_precision['metric_score'], 3))\n",
        "rerank_bert_title_metrics_precision.append(round(cisi_res_rerank_title_precision['metric_score'], 3))\n",
        "rerank_bert_title_metrics_precision.append(round(cran_res_rerank_title_precision['metric_score'], 3))\n",
        "rerank_bert_title_metrics_precision.append(round(lisa_res_rerank_title_precision['metric_score'], 3))\n",
        "\n",
        "\n",
        "labels = ['ADI','CACM', 'CISI','Cranfield', 'LISA']\n",
        "\n",
        "x = np.arange(len(labels))*2.5  # the label locations\n",
        "\n",
        "width = 0.5  # the width of the bars\n",
        "\n",
        "# orange, green, turquoise, black, red, yellow, white\n",
        "pragma_colors =  ['#ffb900', '#8cab13', '#22ab82', '#242526', '#cc0000',  '#ffcc00', '#ffffff']\n",
        "\n",
        "fig, ax = plt.subplots()\n",
        "rects1 = ax.bar(x - 2*width, match_metrics_precision_text_title , width, label='With multi-match query on title and text field', color=pragma_colors[0])\n",
        "rects2 = ax.bar(x - width, complex_rerank_bert_metrics_precision , width, label='With complex re-ranking', color=pragma_colors[3])\n",
        "rects3 = ax.bar(x, rerank_bert_text_metrics_precision, width, label='With re-ranking on text field', color=pragma_colors[1])\n",
        "rects4 = ax.bar(x + width, rerank_bert_title_metrics_precision, width, label='With re-ranking on title field', color=pragma_colors[2])\n",
        "\n",
        "# Add some text for labels, title and custom x-axis tick labels, etc.\n",
        "ax.set_ylabel('metric scores')\n",
        "ax.set_title('Complex re-ranking precision scores by corpus')\n",
        "ax.set_xticks(x)\n",
        "ax.set_xticklabels(labels)\n",
        "ax.legend()\n",
        "\n",
        "def autolabel(rects):\n",
        "    \"\"\"Attach a text label above each bar in *rects*, displaying its height.\"\"\"\n",
        "    for rect in rects:\n",
        "        height = rect.get_height()\n",
        "        ax.annotate('{}'.format(height),\n",
        "                    xy=(rect.get_x() + rect.get_width() / 2, height),\n",
        "                    xytext=(0, 3),  # 3 points vertical offset\n",
        "                    textcoords=\"offset points\",\n",
        "                    ha='center', va='bottom')\n",
        "\n",
        "autolabel(rects1)\n",
        "autolabel(rects2)\n",
        "autolabel(rects3)\n",
        "autolabel(rects4)\n",
        "\n",
        "fig.tight_layout()\n",
        "fig.set_figwidth(16)\n",
        "fig.set_figheight(10)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CsGWahQ71qRu"
      },
      "source": [
        "### F-Score"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1NdPTM1Q1qRu"
      },
      "source": [
        "**F-Score on the 'title' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 647
        },
        "id": "t2tbJZa21qRu",
        "outputId": "6bf10a04-47d7-4964-df26-b4bdf4eb4ba4"
      },
      "source": [
        "import matplotlib\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "def f_score(recall,precision):\n",
        "  if recall or precision != 0:\n",
        "    fscore = (2*(recall*precision)/(recall+precision))\n",
        "    return fscore\n",
        "\n",
        "match_metrics_fscore_title = []\n",
        "match_metrics_fscore_title.append(round(f_score(adi_res_match_recall_title['metric_score'], adi_res_match_precision_title['metric_score']),3))\n",
        "match_metrics_fscore_title.append(round(f_score(cacm_res_match_recall_title['metric_score'], cacm_res_match_precision_title['metric_score']),3))\n",
        "match_metrics_fscore_title.append(round(f_score(cisi_res_match_recall_title['metric_score'], cisi_res_match_precision_title['metric_score']),3))\n",
        "match_metrics_fscore_title.append(round(f_score(cran_res_match_recall_title['metric_score'], cran_res_match_precision_title['metric_score']),3))\n",
        "match_metrics_fscore_title.append(round(f_score(lisa_res_match_recall_title['metric_score'], lisa_res_match_precision_title['metric_score']),3))\n",
        "\n",
        "bert_title_metrics_fscore = []\n",
        "bert_title_metrics_fscore.append(round(f_score(adi_res_bert_title_recall['metric_score'], adi_res_bert_title_precision['metric_score']),3))\n",
        "bert_title_metrics_fscore.append(round(f_score(cacm_res_bert_title_recall['metric_score'], cacm_res_bert_title_precision['metric_score']),3))\n",
        "bert_title_metrics_fscore.append(round(f_score(cisi_res_bert_title_recall['metric_score'], cisi_res_bert_title_precision['metric_score']),3))\n",
        "bert_title_metrics_fscore.append(round(f_score(cran_res_bert_title_recall['metric_score'], cran_res_bert_title_precision['metric_score']),3))\n",
        "bert_title_metrics_fscore.append(round(f_score(lisa_res_bert_title_recall['metric_score'], lisa_res_bert_title_precision['metric_score']),3))\n",
        "\n",
        "rerank_bert_title_metrics_fscore = []\n",
        "rerank_bert_title_metrics_fscore.append(round(f_score(adi_res_rerank_title_recall['metric_score'], adi_res_rerank_title_precision['metric_score']),3))\n",
        "rerank_bert_title_metrics_fscore.append(round(f_score(cacm_res_rerank_title_recall['metric_score'], cacm_res_rerank_title_precision['metric_score']),3))\n",
        "rerank_bert_title_metrics_fscore.append(round(f_score(cisi_res_rerank_title_recall['metric_score'], cisi_res_rerank_title_precision['metric_score']),3))\n",
        "rerank_bert_title_metrics_fscore.append(round(f_score(cran_res_rerank_title_recall['metric_score'], cran_res_rerank_title_precision['metric_score']),3))\n",
        "rerank_bert_title_metrics_fscore.append(round(f_score(lisa_res_rerank_title_recall['metric_score'], lisa_res_rerank_title_precision['metric_score']),3))\n",
        "\n",
        "\n",
        "labels = ['ADI','CACM', 'CISI','Cranfield', 'LISA']\n",
        "\n",
        "x = np.arange(len(labels))*2.5  # the label locations\n",
        "\n",
        "width = 0.5  # the width of the bars\n",
        "\n",
        "# orange, green, turquoise, black, red, yellow, white\n",
        "pragma_colors =  ['#ffb900', '#8cab13', '#22ab82', '#242526', '#cc0000',  '#ffcc00', '#ffffff']\n",
        "\n",
        "fig, ax = plt.subplots()\n",
        "rects1 = ax.bar(x - width, match_metrics_fscore_title , width, label='With multi-match query on title', color=pragma_colors[0])\n",
        "rects2 = ax.bar(x , bert_title_metrics_fscore, width, label='With BERT on title field', color=pragma_colors[1])\n",
        "rects3 = ax.bar(x + width, rerank_bert_title_metrics_fscore, width, label='With re-ranking on title field', color=pragma_colors[3])\n",
        "\n",
        "# Add some text for labels, title and custom x-axis tick labels, etc.\n",
        "ax.set_ylabel('metric scores')\n",
        "ax.set_title('F1-Scores on title field by corpus')\n",
        "ax.set_xticks(x)\n",
        "ax.set_xticklabels(labels)\n",
        "ax.legend()\n",
        "\n",
        "\n",
        "def autolabel(rects):\n",
        "    \"\"\"Attach a text label above each bar in *rects*, displaying its height.\"\"\"\n",
        "    for rect in rects:\n",
        "        height = rect.get_height()\n",
        "        ax.annotate('{}'.format(height),\n",
        "                    xy=(rect.get_x() + rect.get_width() / 2, height),\n",
        "                    xytext=(0, 3),  # 3 points vertical offset\n",
        "                    textcoords=\"offset points\",\n",
        "                    ha='center', va='bottom')\n",
        "\n",
        "\n",
        "autolabel(rects1)\n",
        "autolabel(rects2)\n",
        "autolabel(rects3)\n",
        "\n",
        "fig.tight_layout()\n",
        "fig.set_figwidth(16)\n",
        "fig.set_figheight(10)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LlH4YTpl1qRv"
      },
      "source": [
        "**F-Score on the 'text' field**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 531
        },
        "id": "slRMyxKH1qRw",
        "outputId": "2712a894-a466-49a4-d315-0512519084c3"
      },
      "source": [
        "import matplotlib\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "def f_score(recall,precision):\n",
        "  if recall or precision != 0:\n",
        "    fscore = (2*(recall*precision)/(recall+precision))\n",
        "    return fscore\n",
        "\n",
        "match_metrics_fscore_text = []\n",
        "match_metrics_fscore_text.append(round(f_score(adi_res_match_recall_text['metric_score'], adi_res_match_precision_text['metric_score']),3))\n",
        "match_metrics_fscore_text.append(round(f_score(cacm_res_match_recall_text['metric_score'], cacm_res_match_precision_text['metric_score']),3))\n",
        "match_metrics_fscore_text.append(round(f_score(cisi_res_match_recall_text['metric_score'], cisi_res_match_precision_text['metric_score']),3))\n",
        "match_metrics_fscore_text.append(round(f_score(cran_res_match_recall_text['metric_score'], cran_res_match_precision_text['metric_score']),3))\n",
        "match_metrics_fscore_text.append(round(f_score(lisa_res_match_recall_text['metric_score'], lisa_res_match_precision_text['metric_score']),3))\n",
        "match_metrics_fscore_text.append(round(f_score(med_res_match_recall_text['metric_score'], med_res_match_precision_text['metric_score']),3))\n",
        "match_metrics_fscore_text.append(round(f_score(npl_res_match_recall_text['metric_score'], npl_res_match_precision_text['metric_score']),3))\n",
        "match_metrics_fscore_text.append(round(f_score(time_res_match_recall_text['metric_score'], time_res_match_precision_text['metric_score']),3))\n",
        "\n",
        "bert_text_metrics_fscore = []\n",
        "bert_text_metrics_fscore.append(round(f_score(adi_res_bert_text_recall['metric_score'], adi_res_bert_text_precision['metric_score']),3))\n",
        "bert_text_metrics_fscore.append(round(f_score(cacm_res_bert_text_recall['metric_score'], cacm_res_bert_text_precision['metric_score']),3))\n",
        "bert_text_metrics_fscore.append(round(f_score(cisi_res_bert_text_recall['metric_score'], cisi_res_bert_text_precision['metric_score']),3))\n",
        "bert_text_metrics_fscore.append(round(f_score(cran_res_bert_text_recall['metric_score'], cran_res_bert_text_precision['metric_score']),3))\n",
        "bert_text_metrics_fscore.append(round(f_score(lisa_res_bert_text_recall['metric_score'], lisa_res_bert_text_precision['metric_score']),3))\n",
        "bert_text_metrics_fscore.append(round(f_score(med_res_bert_text_recall['metric_score'], med_res_bert_text_precision['metric_score']),3))\n",
        "bert_text_metrics_fscore.append(round(f_score(npl_res_bert_text_recall['metric_score'], npl_res_bert_text_precision['metric_score']),3))\n",
        "bert_text_metrics_fscore.append(round(f_score(time_res_bert_text_recall['metric_score'], time_res_bert_text_precision['metric_score']),3))\n",
        "\n",
        "rerank_bert_text_metrics_fscore = []\n",
        "rerank_bert_text_metrics_fscore.append(round(f_score(adi_res_rerank_text_recall['metric_score'], adi_res_rerank_text_precision['metric_score']),3))\n",
        "rerank_bert_text_metrics_fscore.append(round(f_score(cacm_res_rerank_text_recall['metric_score'], cacm_res_rerank_text_precision['metric_score']),3))\n",
        "rerank_bert_text_metrics_fscore.append(round(f_score(cisi_res_rerank_text_recall['metric_score'], cisi_res_rerank_text_precision['metric_score']),3))\n",
        "rerank_bert_text_metrics_fscore.append(round(f_score(cran_res_rerank_text_recall['metric_score'], cran_res_rerank_text_precision['metric_score']),3))\n",
        "rerank_bert_text_metrics_fscore.append(round(f_score(lisa_res_rerank_text_recall['metric_score'], lisa_res_rerank_text_precision['metric_score']),3))\n",
        "rerank_bert_text_metrics_fscore.append(round(f_score(med_res_rerank_text_recall['metric_score'], med_res_rerank_text_precision['metric_score']),3))\n",
        "rerank_bert_text_metrics_fscore.append(round(f_score(npl_res_rerank_text_recall['metric_score'], npl_res_rerank_text_precision['metric_score']),3))\n",
        "rerank_bert_text_metrics_fscore.append(round(f_score(time_res_rerank_text_recall['metric_score'], time_res_rerank_text_precision['metric_score']),3))\n",
        "\n",
        "labels = ['ADI','CACM', 'CISI','Cranfield', 'LISA', 'Medline', 'NPL', 'Time']\n",
        "\n",
        "x = np.arange(len(labels))*2.5  # the label locations\n",
        "\n",
        "width = 0.6  # the width of the bars\n",
        "\n",
        "# orange, green, turquoise, black, red, yellow, white\n",
        "pragma_colors =  ['#ffb900', '#8cab13', '#22ab82', '#242526', '#cc0000',  '#ffcc00', '#ffffff']\n",
        "\n",
        "fig, ax = plt.subplots()\n",
        "rects1 = ax.bar(x - width, match_metrics_fscore_text , width, label='With multi-match query on text', color=pragma_colors[0])\n",
        "rects2 = ax.bar(x , bert_text_metrics_fscore, width, label='With BERT on text field', color=pragma_colors[1])\n",
        "rects3 = ax.bar(x + width,rerank_bert_text_metrics_fscore, width, label='With re-ranking on text field', color=pragma_colors[3])\n",
        "\n",
        "# Add some text for labels, title and custom x-axis tick labels, etc.\n",
        "ax.set_ylabel('metric scores')\n",
        "ax.set_title('F1-Scores on text field by corpus')\n",
        "ax.set_xticks(x)\n",
        "ax.set_xticklabels(labels)\n",
        "ax.legend()\n",
        "\n",
        "\n",
        "def autolabel(rects):\n",
        "    \"\"\"Attach a text label above each bar in *rects*, displaying its height.\"\"\"\n",
        "    for rect in rects:\n",
        "        height = rect.get_height()\n",
        "        ax.annotate('{}'.format(height),\n",
        "                    xy=(rect.get_x() + rect.get_width() / 2, height),\n",
        "                    xytext=(0, 3),  # 3 points vertical offset\n",
        "                    textcoords=\"offset points\",\n",
        "                    ha='center', va='bottom')\n",
        "\n",
        "\n",
        "autolabel(rects1)\n",
        "autolabel(rects2)\n",
        "autolabel(rects3)\n",
        "\n",
        "fig.tight_layout()\n",
        "fig.set_figwidth(16)\n",
        "fig.set_figheight(8)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x576 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "BvqJkRjw1qRx"
      },
      "source": [
        "**F-score on a complex re-ranking query**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 647
        },
        "id": "tsmAAaWR1qRx",
        "outputId": "bdf9c011-26ec-4e4b-a9e5-de575fe7777f"
      },
      "source": [
        "import matplotlib\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "\n",
        "def f_score(recall,precision):\n",
        "  if recall or precision != 0:\n",
        "    fscore = (2*(recall*precision)/(recall+precision))\n",
        "    return fscore\n",
        "\n",
        "match_metrics_fscore_text_title = []\n",
        "match_metrics_fscore_text_title.append((round(f_score(adi_res_match_recall_text_title['metric_score'], adi_res_match_precision_text_title['metric_score']),3)))\n",
        "match_metrics_fscore_text_title.append((round(f_score(cacm_res_match_recall_text_title['metric_score'], cacm_res_match_precision_text_title['metric_score']),3)))\n",
        "match_metrics_fscore_text_title.append((round(f_score(cisi_res_match_recall_text_title['metric_score'], cisi_res_match_precision_text_title['metric_score']),3)))\n",
        "match_metrics_fscore_text_title.append((round(f_score(cran_res_match_recall_text_title['metric_score'], cran_res_match_precision_text_title['metric_score']),3)))\n",
        "match_metrics_fscore_text_title.append((round(f_score(lisa_res_match_recall_text_title['metric_score'], lisa_res_match_precision_text_title['metric_score']),3)))\n",
        "\n",
        "complex_rerank_bert_metrics_fscore = []\n",
        "complex_rerank_bert_metrics_fscore.append((round(f_score(adi_res_complex_rerank_recall['metric_score'], adi_res_complex_rerank_precision['metric_score']),3)))\n",
        "complex_rerank_bert_metrics_fscore.append((round(f_score(cacm_res_complex_rerank_recall['metric_score'], cacm_res_complex_rerank_precision['metric_score']),3)))\n",
        "complex_rerank_bert_metrics_fscore.append((round(f_score(cisi_res_complex_rerank_recall['metric_score'], cisi_res_complex_rerank_precision['metric_score']),3)))\n",
        "complex_rerank_bert_metrics_fscore.append((round(f_score(cran_res_complex_rerank_recall['metric_score'], cran_res_complex_rerank_precision['metric_score']),3)))\n",
        "complex_rerank_bert_metrics_fscore.append((round(f_score(lisa_res_complex_rerank_recall['metric_score'], lisa_res_complex_rerank_precision['metric_score']),3)))\n",
        "\n",
        "rerank_bert_text_metrics_fscore = []\n",
        "rerank_bert_text_metrics_fscore.append((round(f_score(adi_res_rerank_text_recall['metric_score'], adi_res_rerank_text_precision['metric_score']),3)))\n",
        "rerank_bert_text_metrics_fscore.append((round(f_score(cacm_res_rerank_text_recall['metric_score'], cacm_res_rerank_text_precision['metric_score']),3)))\n",
        "rerank_bert_text_metrics_fscore.append((round(f_score(cisi_res_rerank_text_recall['metric_score'], cisi_res_rerank_text_precision['metric_score']),3)))\n",
        "rerank_bert_text_metrics_fscore.append((round(f_score(cran_res_rerank_text_recall['metric_score'], cran_res_rerank_text_precision['metric_score']),3)))\n",
        "rerank_bert_text_metrics_fscore.append((round(f_score(lisa_res_rerank_text_recall['metric_score'], lisa_res_rerank_text_precision['metric_score']),3)))\n",
        "\n",
        "rerank_bert_title_metrics_fscore = []\n",
        "rerank_bert_title_metrics_fscore.append((round(f_score(adi_res_rerank_title_recall['metric_score'], adi_res_rerank_title_precision['metric_score']),3)))\n",
        "rerank_bert_title_metrics_fscore.append((round(f_score(cacm_res_rerank_title_recall['metric_score'], cacm_res_rerank_title_precision['metric_score']),3)))\n",
        "rerank_bert_title_metrics_fscore.append((round(f_score(cisi_res_rerank_title_recall['metric_score'], cisi_res_rerank_title_precision['metric_score']),3)))\n",
        "rerank_bert_title_metrics_fscore.append((round(f_score(cran_res_rerank_title_recall['metric_score'], cran_res_rerank_title_precision['metric_score']),3)))\n",
        "rerank_bert_title_metrics_fscore.append((round(f_score(lisa_res_rerank_title_recall['metric_score'], lisa_res_rerank_title_precision['metric_score']),3)))\n",
        "\n",
        "\n",
        "labels = ['ADI','CACM', 'CISI','Cranfield', 'LISA']\n",
        "\n",
        "x = np.arange(len(labels))*2.5  # the label locations\n",
        "\n",
        "width = 0.5  # the width of the bars\n",
        "\n",
        "# orange, green, turquoise, black, red, yellow, white\n",
        "pragma_colors =  ['#ffb900', '#8cab13', '#22ab82', '#242526', '#cc0000',  '#ffcc00', '#ffffff']\n",
        "\n",
        "fig, ax = plt.subplots()\n",
        "rects1 = ax.bar(x - 2*width, match_metrics_fscore_text_title , width, label='With multi-match query on title and text field', color=pragma_colors[0])\n",
        "rects2 = ax.bar(x - width, complex_rerank_bert_metrics_fscore , width, label='With complex re-ranking', color=pragma_colors[3])\n",
        "rects3 = ax.bar(x, rerank_bert_text_metrics_fscore, width, label='With re-ranking on text field', color=pragma_colors[1])\n",
        "rects4 = ax.bar(x + width, rerank_bert_title_metrics_fscore, width, label='With re-ranking on title field', color=pragma_colors[2])\n",
        "\n",
        "# Add some text for labels, title and custom x-axis tick labels, etc.\n",
        "ax.set_ylabel('metric scores')\n",
        "ax.set_title('Complex re-ranking F1-Scores by corpus')\n",
        "ax.set_xticks(x)\n",
        "ax.set_xticklabels(labels)\n",
        "ax.legend()\n",
        "\n",
        "\n",
        "def autolabel(rects):\n",
        "    \"\"\"Attach a text label above each bar in *rects*, displaying its height.\"\"\"\n",
        "    for rect in rects:\n",
        "        height = rect.get_height()\n",
        "        ax.annotate('{}'.format(height),\n",
        "                    xy=(rect.get_x() + rect.get_width() / 2, height),\n",
        "                    xytext=(0, 3),  # 3 points vertical offset\n",
        "                    textcoords=\"offset points\",\n",
        "                    ha='center', va='bottom')\n",
        "\n",
        "\n",
        "autolabel(rects1)\n",
        "autolabel(rects2)\n",
        "autolabel(rects3)\n",
        "autolabel(rects4)\n",
        "\n",
        "fig.tight_layout()\n",
        "fig.set_figwidth(16)\n",
        "fig.set_figheight(10)\n",
        "\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Ikgcrccon7a1"
      },
      "source": [
        "# Read More\n",
        "Details on this experiment can be found on our [website](https://pragmalingu.de/docs/experiments/experiment2)."
      ]
    }
  ]
}