LSTM-UCISpam.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 238,
   "metadata": {
    "collapsed": true,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.268645223Z",
     "start_time": "2023-07-21T18:27:36.240150494Z"
    }
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from torch import nn\n",
    "import os\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 239,
   "outputs": [],
   "source": [
    "# define parameters\n",
    "SEED = 1234\n",
    "BATCH_SIZE = 64"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.271484973Z",
     "start_time": "2023-07-21T18:27:36.241816979Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Data Loading and Preprocessing"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 240,
   "outputs": [
    {
     "data": {
      "text/plain": "  label                                            message\n0   ham  Go until jurong point, crazy.. Available only ...\n1   ham                      Ok lar... Joking wif u oni...\n2  spam  Free entry in 2 a wkly comp to win FA Cup fina...\n3   ham  U dun say so early hor... U c already then say...\n4   ham  Nah I don't think he goes to usf, he lives aro...",
      "text/html": "<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>label</th>\n      <th>message</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>ham</td>\n      <td>Go until jurong point, crazy.. Available only ...</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>ham</td>\n      <td>Ok lar... Joking wif u oni...</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>spam</td>\n      <td>Free entry in 2 a wkly comp to win FA Cup fina...</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>ham</td>\n      <td>U dun say so early hor... U c already then say...</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>ham</td>\n      <td>Nah I don't think he goes to usf, he lives aro...</td>\n    </tr>\n  </tbody>\n</table>\n</div>"
     },
     "execution_count": 240,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "url = 'https://raw.githubusercontent.com/ShresthaSudip/SMS_Spam_Detection_DNN_LSTM_BiLSTM/master/SMSSpamCollection'\n",
    "messages = pd.read_csv(url, sep ='\\t',names=[\"label\", \"message\"])\n",
    "messages.head()"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.456565718Z",
     "start_time": "2023-07-21T18:27:36.249327981Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 241,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "      message                                                               \n",
      "        count unique                                                top freq\n",
      "label                                                                       \n",
      "ham      4825   4516                             Sorry, I'll call later   30\n",
      "spam      747    653  Please call our customer service representativ...    4\n"
     ]
    }
   ],
   "source": [
    "print(messages.groupby(\"label\").describe())"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.456721535Z",
     "start_time": "2023-07-21T18:27:36.455679411Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 242,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 1200x800 with 1 Axes>",
      "image/png": "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"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "messages[\"Message Length\"]=messages[\"message\"].apply(len)\n",
    "fig=plt.figure(figsize=(12,8))\n",
    "sns.histplot(\n",
    "    x=messages[\"Message Length\"],\n",
    "    hue=messages[\"label\"]\n",
    ")\n",
    "plt.title(\"ham & spam messege length comparision\")\n",
    "plt.show()"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.600435358Z",
     "start_time": "2023-07-21T18:27:36.455775733Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 243,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Ham Messege Length Description:\n",
      " count    4825.000000\n",
      "mean       71.482487\n",
      "std        58.440652\n",
      "min         2.000000\n",
      "25%        33.000000\n",
      "50%        52.000000\n",
      "75%        93.000000\n",
      "max       910.000000\n",
      "Name: Message Length, dtype: float64\n",
      "************************************\n",
      "Spam Message Length Description:\n",
      " count    747.000000\n",
      "mean     138.670683\n",
      "std       28.873603\n",
      "min       13.000000\n",
      "25%      133.000000\n",
      "50%      149.000000\n",
      "75%      157.000000\n",
      "max      223.000000\n",
      "Name: Message Length, dtype: float64\n"
     ]
    }
   ],
   "source": [
    "ham_desc=messages[messages[\"label\"]==\"ham\"][\"Message Length\"].describe()\n",
    "spam_desc=messages[messages[\"label\"]==\"spam\"][\"Message Length\"].describe()\n",
    "\n",
    "print(\"Ham Messege Length Description:\\n\",ham_desc)\n",
    "print(\"************************************\")\n",
    "print(\"Spam Message Length Description:\\n\",spam_desc)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.699878931Z",
     "start_time": "2023-07-21T18:27:36.601449342Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 244,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.countplot(\n",
    "    data=messages,\n",
    "    x=\"label\"\n",
    ")\n",
    "plt.title(\"ham vs spam\")\n",
    "plt.show()"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.700380569Z",
     "start_time": "2023-07-21T18:27:36.644462996Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "### Data Cleaning (undersampling for imbalanced data)"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 245,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "df shape:  (1494, 3)\n"
     ]
    },
    {
     "data": {
      "text/plain": "spam    747\nham     747\nName: label, dtype: int64"
     },
     "execution_count": 245,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#compute the length of majority & minority class\n",
    "minority_len=len(messages[messages[\"label\"]==\"spam\"])\n",
    "majority_len=len(messages[messages[\"label\"]==\"ham\"])\n",
    "\n",
    "#store the indices of majority and minority class\n",
    "minority_indices=messages[messages[\"label\"]==\"spam\"].index\n",
    "majority_indices=messages[messages[\"label\"]==\"ham\"].index\n",
    "\n",
    "#generate new majority indices from the total majority_indices\n",
    "#with size equal to minority class length so we obtain equivalent number of indices length\n",
    "random_majority_indices=np.random.choice(\n",
    "    majority_indices,\n",
    "    size=minority_len,\n",
    "    replace=False\n",
    ")\n",
    "\n",
    "#concatenate the two indices to obtain indices of new dataframe\n",
    "undersampled_indices=np.concatenate([minority_indices,random_majority_indices])\n",
    "\n",
    "#create df using new indices\n",
    "df=messages.loc[undersampled_indices]\n",
    "\n",
    "#shuffle the sample\n",
    "df=df.sample(frac=1)\n",
    "\n",
    "#reset the index as its all mixed\n",
    "df=df.reset_index()\n",
    "\n",
    "#drop the older index\n",
    "df=df.drop(\n",
    "    columns=[\"index\"],\n",
    ")\n",
    "print(\"df shape: \", df.shape)\n",
    "df['label'].value_counts()"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.700493576Z",
     "start_time": "2023-07-21T18:27:36.688584034Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 246,
   "outputs": [],
   "source": [
    "df[\"Label\"]=df[\"label\"].map(\n",
    "    {\n",
    "        \"ham\":0,\n",
    "        \"spam\":1\n",
    "    }\n",
    ")"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.700528743Z",
     "start_time": "2023-07-21T18:27:36.688771233Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 247,
   "outputs": [],
   "source": [
    "# save df to csv\n",
    "df.to_csv(\"df.csv\",index=False)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:36.700560856Z",
     "start_time": "2023-07-21T18:27:36.688873860Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "### Data Processing and Tokenization"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 248,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'label': 'spam', 'text': ['Please', 'call', 'our', 'customer', 'service', 'representative', 'on', 'FREEPHONE', '0808', '145', '4742', 'between', '9am-11pm', 'as', 'you', 'have', 'WON', 'a', 'guaranteed', '£', '1000', 'cash', 'or', '£', '5000', 'prize', '!']}\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "from torchtext import data\n",
    "import warnings as wrn\n",
    "import spacy\n",
    "wrn.filterwarnings(\"ignore\")\n",
    "\n",
    "spacy_en = spacy.load('en_core_web_sm')\n",
    "text = data.Field(tokenize=lambda text: [tok.text for tok in spacy_en(text)], batch_first=True, include_lengths=True, unk_token='<unk>')\n",
    "label = data.LabelField(dtype = torch.float,batch_first=True)\n",
    "\n",
    "fields = [(\"label\",label),('text',text)]\n",
    "\n",
    "# load data from df\n",
    "training_data=data.TabularDataset(\n",
    "    path=\"df.csv\",\n",
    "    format=\"csv\",\n",
    "    fields=fields,\n",
    "    skip_header=True,\n",
    ")\n",
    "print(vars(training_data.examples[1]))"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "is_executing": true
    },
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:40.945523966Z",
     "start_time": "2023-07-21T18:27:36.688988815Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 249,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Size of text vocab:  1378\n",
      "Size of label vocab:  2\n"
     ]
    }
   ],
   "source": [
    "import random\n",
    "train_data, test_data = training_data.split(split_ratio=0.8, random_state=random.seed(SEED))\n",
    "text.build_vocab(train_data, min_freq=3, vectors=\"glove.6B.100d\")\n",
    "label.build_vocab(train_data)\n",
    "\n",
    "torch.save(text.vocab, 'text_vocab.pkl')\n",
    "torch.save(label.vocab, 'label_vocab.pkl')\n",
    "\n",
    "print(\"Size of text vocab: \", len(text.vocab))\n",
    "print(\"Size of label vocab: \", len(label.vocab))"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:41.089496044Z",
     "start_time": "2023-07-21T18:27:40.949681244Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 250,
   "outputs": [
    {
     "data": {
      "text/plain": "[('.', 1279),\n ('to', 664),\n ('!', 489),\n (',', 476),\n ('a', 400),\n ('you', 359),\n ('?', 289),\n ('I', 285),\n ('the', 278),\n ('£', 233)]"
     },
     "execution_count": 250,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "text.vocab.freqs.most_common(10)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:41.092312302Z",
     "start_time": "2023-07-21T18:27:41.090874215Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Model Building & Data Loading"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 251,
   "outputs": [],
   "source": [
    "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
    "\n",
    "train_iterator, test_iterator = data.BucketIterator.splits(\n",
    "    (train_data, test_data),\n",
    "    batch_size = BATCH_SIZE,\n",
    "    sort_key = lambda x: len(x.text),\n",
    "    sort_within_batch=True,\n",
    "    device = device)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:41.099546991Z",
     "start_time": "2023-07-21T18:27:41.092419476Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 252,
   "outputs": [],
   "source": [
    "import torch.nn as nn\n",
    "\n",
    "class LSTM(nn.Module):\n",
    "    def __init__(self, vocab_size, embedding_dim, hidden_dim, output_dim, n_layers, bidirectional, dropout):\n",
    "        super().__init__()\n",
    "        self.embedding = nn.Embedding(vocab_size, embedding_dim)\n",
    "        # lstm layer -> output_dim\n",
    "        self.lstm = nn.LSTM(embedding_dim, hidden_dim, num_layers=n_layers, bidirectional=bidirectional, dropout=dropout)\n",
    "        # dense layer -> prediction\n",
    "        self.fc = nn.Linear(hidden_dim * 2, output_dim)\n",
    "        # dropout layer\n",
    "        self.dropout = nn.Dropout(dropout)\n",
    "        # activation function\n",
    "        self.act = nn.Sigmoid()\n",
    "\n",
    "    def forward(self, text, text_lengths):\n",
    "        embedded = self.dropout(self.embedding(text))\n",
    "        # pack sequence\n",
    "        packed_embedded = nn.utils.rnn.pack_padded_sequence(embedded, text_lengths.cpu(), batch_first=True)\n",
    "        packed_output, (hidden, cell) = self.lstm(packed_embedded)\n",
    "        # unpack sequence\n",
    "        output, output_lengths = nn.utils.rnn.pad_packed_sequence(packed_output, batch_first=True)\n",
    "\n",
    "        # Use the last hidden states of the LSTM layers from both directions\n",
    "        hidden = self.dropout(torch.cat((hidden[-2, :, :], hidden[-1, :, :]), dim=1))\n",
    "\n",
    "        # Note: hidden has shape (num_layers*num_directions, batch_size, hidden_dim)\n",
    "        # If you want to use only the last layer's hidden state, you can modify it like this:\n",
    "        # hidden = self.dropout(hidden[-1, :, :])\n",
    "\n",
    "        return self.act(self.fc(hidden))\n"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:41.144443721Z",
     "start_time": "2023-07-21T18:27:41.100838467Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 253,
   "outputs": [],
   "source": [
    "# training hyperparameters\n",
    "SIZE_OF_VOCAB = len(text.vocab)\n",
    "EMBEDDING_DIM = 100\n",
    "HIDDEN_DIM = 256\n",
    "OUTPUT_DIM = 1\n",
    "N_LAYERS = 2\n",
    "BIDIRECTIONAL = True\n",
    "DROPOUT = 0.2\n",
    "NUM_EPOCHS = 15"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:41.145270632Z",
     "start_time": "2023-07-21T18:27:41.144588078Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Training"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 254,
   "outputs": [
    {
     "data": {
      "text/plain": "LSTM(\n  (embedding): Embedding(1378, 100)\n  (lstm): LSTM(100, 256, num_layers=2, dropout=0.2, bidirectional=True)\n  (fc): Linear(in_features=512, out_features=1, bias=True)\n  (dropout): Dropout(p=0.2, inplace=False)\n  (act): Sigmoid()\n)"
     },
     "execution_count": 254,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = LSTM(SIZE_OF_VOCAB, EMBEDDING_DIM, HIDDEN_DIM, OUTPUT_DIM, N_LAYERS, BIDIRECTIONAL, DROPOUT)\n",
    "optimizer = torch.optim.Adam(model.parameters())\n",
    "criterion = nn.BCELoss()\n",
    "model = model.to(device)\n",
    "model"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:41.145419337Z",
     "start_time": "2023-07-21T18:27:41.144712803Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 255,
   "outputs": [],
   "source": [
    "def binary_accuracy(preds, y):\n",
    "    # round predictions to the closest integer\n",
    "    rounded_preds = torch.round(preds)\n",
    "    correct = (rounded_preds == y).float() # convert into float for division\n",
    "    acc = correct.sum() / len(correct)\n",
    "    return acc"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:41.145456837Z",
     "start_time": "2023-07-21T18:27:41.144764272Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 256,
   "outputs": [],
   "source": [
    "def train(model, iterator, optimizer, criterion):\n",
    "    epoch_loss = 0\n",
    "    epoch_acc = 0\n",
    "    model.train()\n",
    "    for batch in iterator:\n",
    "        optimizer.zero_grad()\n",
    "        text, text_lengths = batch.text\n",
    "        predictions = model(text, text_lengths).squeeze(1)\n",
    "        loss = criterion(predictions, batch.label)\n",
    "        acc = binary_accuracy(predictions, batch.label)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        epoch_loss += loss.item()\n",
    "        epoch_acc += acc.item()\n",
    "    return epoch_loss / len(iterator), epoch_acc / len(iterator)\n",
    "\n",
    "def evaluate(model, iterator, criterion):\n",
    "    epoch_loss = 0\n",
    "    epoch_acc = 0\n",
    "    model.eval()\n",
    "    with torch.no_grad():\n",
    "        for batch in iterator:\n",
    "            text, text_lengths = batch.text\n",
    "            predictions = model(text, text_lengths).squeeze(1)\n",
    "            loss = criterion(predictions, batch.label)\n",
    "            acc = binary_accuracy(predictions, batch.label)\n",
    "            epoch_loss += loss.item()\n",
    "            epoch_acc += acc.item()\n",
    "    return epoch_loss / len(iterator), epoch_acc / len(iterator)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:27:41.145495931Z",
     "start_time": "2023-07-21T18:27:41.144801284Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 257,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 01\n",
      "\tTrain Loss: 0.672 | Train Acc: 61.73%\n",
      "\tTest Loss: 0.566 | Test Acc: 83.00%\n",
      "Epoch: 02\n",
      "\tTrain Loss: 0.407 | Train Acc: 83.44%\n",
      "\tTest Loss: 0.337 | Test Acc: 85.49%\n",
      "Epoch: 03\n",
      "\tTrain Loss: 0.216 | Train Acc: 91.86%\n",
      "\tTest Loss: 0.314 | Test Acc: 89.39%\n",
      "Epoch: 04\n",
      "\tTrain Loss: 0.178 | Train Acc: 93.26%\n",
      "\tTest Loss: 0.163 | Test Acc: 93.76%\n",
      "Epoch: 05\n",
      "\tTrain Loss: 0.112 | Train Acc: 95.89%\n",
      "\tTest Loss: 0.172 | Test Acc: 93.92%\n",
      "Epoch: 06\n",
      "\tTrain Loss: 0.078 | Train Acc: 98.03%\n",
      "\tTest Loss: 0.212 | Test Acc: 94.22%\n",
      "Epoch: 07\n",
      "\tTrain Loss: 0.075 | Train Acc: 97.37%\n",
      "\tTest Loss: 0.181 | Test Acc: 94.69%\n",
      "Epoch: 08\n",
      "\tTrain Loss: 0.065 | Train Acc: 97.70%\n",
      "\tTest Loss: 0.148 | Test Acc: 95.32%\n",
      "Epoch: 09\n",
      "\tTrain Loss: 0.060 | Train Acc: 97.53%\n",
      "\tTest Loss: 0.136 | Test Acc: 95.16%\n",
      "Epoch: 10\n",
      "\tTrain Loss: 0.060 | Train Acc: 98.07%\n",
      "\tTest Loss: 0.135 | Test Acc: 95.32%\n",
      "Epoch: 11\n",
      "\tTrain Loss: 0.041 | Train Acc: 98.32%\n",
      "\tTest Loss: 0.140 | Test Acc: 95.32%\n",
      "Epoch: 12\n",
      "\tTrain Loss: 0.027 | Train Acc: 99.01%\n",
      "\tTest Loss: 0.127 | Test Acc: 96.25%\n",
      "Epoch: 13\n",
      "\tTrain Loss: 0.017 | Train Acc: 99.67%\n",
      "\tTest Loss: 0.108 | Test Acc: 96.10%\n",
      "Epoch: 14\n",
      "\tTrain Loss: 0.014 | Train Acc: 99.67%\n",
      "\tTest Loss: 0.170 | Test Acc: 95.78%\n",
      "Epoch: 15\n",
      "\tTrain Loss: 0.023 | Train Acc: 98.93%\n",
      "\tTest Loss: 0.178 | Test Acc: 95.47%\n"
     ]
    }
   ],
   "source": [
    "for epoch in range(NUM_EPOCHS):\n",
    "    train_loss, train_acc = train(model, train_iterator, optimizer, criterion)\n",
    "    test_loss, test_acc = evaluate(model, test_iterator, criterion)\n",
    "    print(f'Epoch: {epoch+1:02}')\n",
    "    print(f'\\tTrain Loss: {train_loss:.3f} | Train Acc: {train_acc*100:.2f}%')\n",
    "    print(f'\\tTest Loss: {test_loss:.3f} | Test Acc: {test_acc*100:.2f}%')"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:28:12.586532642Z",
     "start_time": "2023-07-21T18:27:41.144851608Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 258,
   "outputs": [],
   "source": [
    "# save the model\n",
    "torch.save(model.state_dict(), 'LSTM-UCIspam-model.pt')"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:28:12.595596954Z",
     "start_time": "2023-07-21T18:28:12.586692792Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## fine-tuning with our own data"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 259,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Size of text vocab:  1378\n",
      "Epoch: 01\n",
      "\tTrain Loss: 1.242 | Train Acc: 65.22%\n",
      "Epoch: 02\n",
      "\tTrain Loss: 0.642 | Train Acc: 65.22%\n",
      "Epoch: 03\n",
      "\tTrain Loss: 0.188 | Train Acc: 95.65%\n",
      "Epoch: 04\n",
      "\tTrain Loss: 0.394 | Train Acc: 78.26%\n",
      "Epoch: 05\n",
      "\tTrain Loss: 0.463 | Train Acc: 82.61%\n",
      "Epoch: 06\n",
      "\tTrain Loss: 0.145 | Train Acc: 95.65%\n",
      "Epoch: 07\n",
      "\tTrain Loss: 0.084 | Train Acc: 100.00%\n",
      "Epoch: 08\n",
      "\tTrain Loss: 0.073 | Train Acc: 100.00%\n",
      "Epoch: 09\n",
      "\tTrain Loss: 0.067 | Train Acc: 100.00%\n",
      "Epoch: 10\n",
      "\tTrain Loss: 0.082 | Train Acc: 100.00%\n",
      "Epoch: 11\n",
      "\tTrain Loss: 0.075 | Train Acc: 100.00%\n",
      "Epoch: 12\n",
      "\tTrain Loss: 0.101 | Train Acc: 100.00%\n",
      "Epoch: 13\n",
      "\tTrain Loss: 0.080 | Train Acc: 100.00%\n",
      "Epoch: 14\n",
      "\tTrain Loss: 0.078 | Train Acc: 100.00%\n",
      "Epoch: 15\n",
      "\tTrain Loss: 0.078 | Train Acc: 95.65%\n"
     ]
    }
   ],
   "source": [
    "# fine tune with our fine-tune-dataset.csv\n",
    "import pandas as pd\n",
    "ft_df = pd.read_csv(\"fine-tune-dataset.csv\")\n",
    "\n",
    "# convert label from str to int\n",
    "ft_df[\"Label\"]=ft_df[\"label\"].map(\n",
    "    {\n",
    "        \"ham\":0,\n",
    "        \"spam\":1\n",
    "    }\n",
    ")\n",
    "\n",
    "# use all data to train\n",
    "ft_data = data.TabularDataset(\n",
    "    path=\"fine-tune-dataset.csv\",\n",
    "    format=\"csv\",\n",
    "    fields=[(\"label\", label), ('text', text)],  # Reuse the existing fields\n",
    "    skip_header=True,\n",
    ")\n",
    "\n",
    "# load and extent the vocab\n",
    "text.vocab = torch.load('text_vocab.pkl')\n",
    "label.vocab = torch.load('label_vocab.pkl')\n",
    "print(\"Size of text vocab: \", len(text.vocab))\n",
    "\n",
    "\n",
    "# load model\n",
    "model.load_state_dict(torch.load('LSTM-UCIspam-model.pt'))\n",
    "model = model.to(device)\n",
    "\n",
    "# training\n",
    "train_iterator = data.BucketIterator(\n",
    "    ft_data,\n",
    "    batch_size = BATCH_SIZE,\n",
    "    sort_key = lambda x: len(x.text),\n",
    "    sort_within_batch=True,\n",
    "    device = device)\n",
    "\n",
    "\n",
    "for epoch in range(NUM_EPOCHS):\n",
    "    train_loss, train_acc = train(model, train_iterator, optimizer, criterion)\n",
    "    print(f'Epoch: {epoch+1:02}')\n",
    "    print(f'\\tTrain Loss: {train_loss:.3f} | Train Acc: {train_acc*100:.2f}%')\n",
    "\n",
    "\n",
    "# save the model\n",
    "torch.save(model.state_dict(), 'LSTM-fine-tune-model.pt')\n",
    "\n",
    "# Save the tokenizer configuration (including the tokenization function)\n",
    "tokenizer_config = {\n",
    "    'tokenizer_type': 'spacy',\n",
    "    'language_model': 'en_core_web_sm',\n",
    "    'batch_first': True,\n",
    "    'include_lengths': True\n",
    "}\n",
    "\n",
    "# Save the tokenizer configuration to a JSON file\n",
    "import json\n",
    "with open('tokenizer_config.json', 'w') as f:\n",
    "    json.dump(tokenizer_config, f)\n"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:28:13.260445835Z",
     "start_time": "2023-07-21T18:28:12.597346987Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Testing with some examples"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 260,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([1, 15])\n",
      "t1 prediction: 0.9972\n",
      "torch.Size([1, 24])\n",
      "t2 prediction: 0.1015\n",
      "torch.Size([1, 1])\n",
      "t3 prediction: 0.7124\n"
     ]
    }
   ],
   "source": [
    "t1 = \"Limited time offer! Buy one, get one free on all online courses!\"\n",
    "t2 = \"Just finished my final exams! Looking forward to a relaxing summer break. What are some places you recommend for a vacation?\"\n",
    "t3 = \"aabbaac\"\n",
    "\n",
    "def tokenizer(text: str):\n",
    "    return [tok.text for tok in spacy_en(text)]\n",
    "\n",
    "\n",
    "\n",
    "def token2index(token: str):\n",
    "    return text.vocab.stoi[token]\n",
    "\n",
    "def predict(text: str):\n",
    "    # Tokenize the text\n",
    "    tokens = tokenizer(text)\n",
    "    # Convert tokens to indices\n",
    "    indices = [token2index(token) for token in tokens]\n",
    "    # Convert indices to a PyTorch tensor\n",
    "    indices_tensor = torch.LongTensor(indices).unsqueeze(0).to(device)\n",
    "    print (indices_tensor.shape)\n",
    "    # Compute the prediction\n",
    "    prediction = model(indices_tensor, torch.LongTensor([len(indices)]).to(device))\n",
    "    return prediction.item()\n",
    "\n",
    "print(\"t1 prediction: %.4f\" % predict(t1))\n",
    "print(\"t2 prediction: %.4f\" % predict(t2))\n",
    "print(\"t3 prediction: %.4f\" % predict(t3))\n",
    "\n"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:28:13.283928962Z",
     "start_time": "2023-07-21T18:28:13.262346399Z"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Save the model with onnx"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 270,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "================ Diagnostic Run torch.onnx.export version 2.0.1 ================\n",
      "verbose: False, log level: Level.ERROR\n",
      "======================= 0 NONE 0 NOTE 2 WARNING 0 ERROR ========================\n",
      "2 WARNING were not printed due to the log level.\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[W shape_type_inference.cpp:1920] Warning: The shape inference of prim::PackPadded type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function. (function UpdateReliable)\n",
      "[W shape_type_inference.cpp:1920] Warning: The shape inference of prim::PackPadded type is missing, so it may result in wrong shape inference for the exported graph. Please consider adding it in symbolic function. (function UpdateReliable)\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "import torch.onnx as onnx\n",
    "\n",
    "# inputs of shape [1, max_sequence_length]\n",
    "max_sequence_length = 1000\n",
    "\n",
    "sample_input = torch.zeros((1, max_sequence_length), dtype=torch.long), torch.LongTensor([max_sequence_length])\n",
    "\n",
    "# Export the model to ONNX format with dynamic sequence length\n",
    "output_onnx_path = \"spam-lstm.onnx\"\n",
    "onnx.export(\n",
    "    model,\n",
    "    sample_input,\n",
    "    output_onnx_path,\n",
    "    input_names=[\"input\", \"text_lengths\"],\n",
    "    output_names=[\"output\"],\n",
    "    dynamic_axes={\"input\": {0: \"batch_size\", 1: \"sequence_length\"},\n",
    "                  \"text_length\": {0: \"text_length\"}}\n",
    ")\n"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "end_time": "2023-07-21T18:42:39.660065545Z",
     "start_time": "2023-07-21T18:42:39.366138331Z"
    }
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}