BiLSTM_attention_keras.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# KERAS, 시계열, BiLSTM, Attention - 181204\n",
    "https://tykimos.github.io/2017/09/09/Time-series_Numerical_Input_Numerical_Prediction_Model_Recipe/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "# 0. 사용할 패키지 불러오기\n",
    "import numpy as np\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, LSTM, Dropout\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import pickle\n",
    "from collections import defaultdict\n",
    "import re\n",
    "\n",
    "from bs4 import BeautifulSoup\n",
    "\n",
    "import sys\n",
    "import os\n",
    "from nltk import tokenize\n",
    "\n",
    "from keras.preprocessing.text import Tokenizer, text_to_word_sequence\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.utils.np_utils import to_categorical\n",
    "\n",
    "from keras.layers import Embedding, RepeatVector, Permute, Multiply, Lambda, BatchNormalization, LeakyReLU\n",
    "from keras.layers import Dense, Input, Flatten\n",
    "from keras.layers import Conv1D, MaxPooling1D, Embedding, merge, Dropout, LSTM, CuDNNLSTM, GRU, Bidirectional, TimeDistributed\n",
    "from keras.models import Model\n",
    "from keras.callbacks import EarlyStopping, LearningRateScheduler, Callback, ModelCheckpoint\n",
    "\n",
    "from keras.optimizers import Adam, Adadelta, SGD\n",
    "\n",
    "\n",
    "from keras import backend as K\n",
    "from keras.engine.topology import Layer, InputSpec\n",
    "from keras import initializers\n",
    "\n",
    "from tqdm import tqdm\n",
    "import math\n",
    "from matplotlib import pyplot as plt\n",
    "import seaborn as sn\n",
    "from colored import fg, bg, attr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "class AttentionLayer(Layer):\n",
    "    \"\"\"\n",
    "    Attention operation, with a context/query vector, for temporal data.\n",
    "    Supports Masking.\n",
    "    Follows the work of Yang et al. [https://www.cs.cmu.edu/~diyiy/docs/naacl16.pdf]\n",
    "    \"Hierarchical Attention Networks for Document Classification\"\n",
    "    by using a context vector to assist the attention\n",
    "    # Input shape\n",
    "        3D tensor with shape: `(samples, steps, features)`.\n",
    "    # Output shape\n",
    "        2D tensor with shape: `(samples, features)`.\n",
    "    How to use:\n",
    "    Just put it on top of an RNN Layer (GRU/LSTM/SimpleRNN) with return_sequences=True.\n",
    "    The dimensions are inferred based on the output shape of the RNN.\n",
    "    Note: The layer has been tested with Keras 2.0.6\n",
    "    Example:\n",
    "        model.add(LSTM(64, return_sequences=True))\n",
    "        model.add(AttentionWithContext())\n",
    "        # next add a Dense layer (for classification/regression) or whatever...\n",
    "    \"\"\"\n",
    "    def __init__(self, **kwargs):\n",
    "        self.init = initializers.get('glorot_uniform')\n",
    "        super(AttentionLayer, self).__init__(**kwargs)\n",
    "    \n",
    "    def build(self, input_shape):\n",
    "        assert len(input_shape) == 3\n",
    "        \n",
    "        self.W = self.add_weight(name='Attention_Weight',\n",
    "                                 shape=(input_shape[-1], input_shape[-1]),\n",
    "                                 initializer=self.init,\n",
    "                                 trainable=True)\n",
    "        self.b = self.add_weight(name='Attention_Bias',\n",
    "                                 shape=(input_shape[-1], ),\n",
    "                                 initializer=self.init,\n",
    "                                 trainable=True)\n",
    "        self.u = self.add_weight(name='Attention_Context_Vector',\n",
    "                                 shape=(input_shape[-1], 1),\n",
    "                                 initializer=self.init,\n",
    "                                 trainable=True)\n",
    "        super(AttentionLayer, self).build(input_shape)\n",
    "\n",
    "    def compute_mask(self, input, input_mask=None):\n",
    "        # do not pass the mask to the next layers\n",
    "        return None\n",
    "    \n",
    "    def call(self, x):\n",
    "        # refer to the original paper\n",
    "        # link: https://www.cs.cmu.edu/~hovy/papers/16HLT-hierarchical-attention-networks.pdf\n",
    "        \n",
    "        # RNN 구조를 거쳐서 나온 hidden states (x)에 single layer perceptron (tanh activation)\n",
    "        # 적용하여 나온 벡터가 uit \n",
    "        u_it = K.tanh(K.dot(x, self.W) + self.b)\n",
    "        \n",
    "        # uit와 uw (혹은 us) 간의 similarity를 attention으로 사용\n",
    "        # softmax를 통해 attention 값을 확률 분포로 만듬\n",
    "        a_it = K.dot(u_it, self.u)\n",
    "        a_it = K.squeeze(a_it, -1)\n",
    "        a_it = K.softmax(a_it)\n",
    "        \n",
    "        return a_it\n",
    "        \n",
    "    def compute_output_shape(self, input_shape):\n",
    "        return (input_shape[0], input_shape[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_dataset(signal_data, look_back=1):\n",
    "    dataX, dataY = [], []\n",
    "    for i in range(len(signal_data)-look_back):\n",
    "        dataX.append(signal_data[i:(i+look_back), 0])\n",
    "        dataY.append(signal_data[i + look_back, 0])\n",
    "    return np.array(dataX), np.array(dataY)\n",
    "\n",
    "look_back = 40\n",
    "\n",
    "# 1. 데이터셋 생성하기\n",
    "signal_data = np.cos(np.arange(1600)*(20*np.pi/1000))[:,None]\n",
    "\n",
    "# 데이터 전처리\n",
    "scaler = MinMaxScaler(feature_range=(0, 1))\n",
    "signal_data = scaler.fit_transform(signal_data)\n",
    "\n",
    "# 데이터 분리\n",
    "train = signal_data[0:800]\n",
    "val = signal_data[800:1200]\n",
    "test = signal_data[1200:]\n",
    "\n",
    "# 데이터셋 생성\n",
    "x_train, y_train = create_dataset(train, look_back)\n",
    "x_val, y_val = create_dataset(val, look_back)\n",
    "x_test, y_test = create_dataset(test, look_back)\n",
    "\n",
    "# 데이터셋 전처리\n",
    "x_train = np.reshape(x_train, (x_train.shape[0], x_train.shape[1], 1))\n",
    "x_val = np.reshape(x_val, (x_val.shape[0], x_val.shape[1], 1))\n",
    "x_test = np.reshape(x_test, (x_test.shape[0], x_test.shape[1], 1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensor(\"attention_layer_4/Softmax:0\", shape=(?, 40), dtype=float32)\n",
      "Train on 760 samples, validate on 360 samples\n",
      "Epoch 1/10\n",
      "760/760 [==============================] - 2s 3ms/step - loss: 0.1749 - val_loss: 0.1301\n",
      "Epoch 2/10\n",
      "760/760 [==============================] - 1s 2ms/step - loss: 0.1213 - val_loss: 0.1149\n",
      "Epoch 3/10\n",
      "760/760 [==============================] - 1s 2ms/step - loss: 0.0944 - val_loss: 0.0675\n",
      "Epoch 4/10\n",
      "760/760 [==============================] - 1s 2ms/step - loss: 0.0251 - val_loss: 0.0061\n",
      "Epoch 5/10\n",
      "760/760 [==============================] - 1s 2ms/step - loss: 0.0035 - val_loss: 0.0012\n",
      "Epoch 6/10\n",
      "760/760 [==============================] - 1s 2ms/step - loss: 6.1331e-04 - val_loss: 3.7459e-04\n",
      "Epoch 7/10\n",
      "760/760 [==============================] - 1s 2ms/step - loss: 1.4554e-04 - val_loss: 7.3269e-05\n",
      "Epoch 8/10\n",
      "760/760 [==============================] - 1s 2ms/step - loss: 6.0771e-05 - val_loss: 4.3570e-05\n",
      "Epoch 9/10\n",
      "760/760 [==============================] - 1s 2ms/step - loss: 4.4625e-05 - val_loss: 4.2504e-05\n",
      "Epoch 10/10\n",
      "760/760 [==============================] - 1s 2ms/step - loss: 4.2591e-05 - val_loss: 3.0230e-05\n"
     ]
    }
   ],
   "source": [
    "lstm_dim = 50\n",
    "\n",
    "input_data = Input(shape=(40,1))\n",
    "bilstm = Bidirectional(LSTM(lstm_dim, return_sequences=True))(input_data)  # lstm입력은 (N, X, Y) 3차원이어여한다\n",
    "# bilstm = LSTM(2*lstm_dim, return_sequences=True)(input_data)  # lstm입력은 (N, X, Y) 3차원이어여한다\n",
    "# bilstm_output = Dense(1)(bilstm)\n",
    "\n",
    "attention_layer = AttentionLayer()(bilstm)\n",
    "print(attention_layer)\n",
    "\n",
    "repeated_word_attention = RepeatVector(lstm_dim * 2)(attention_layer)\n",
    "repeated_word_attention = Permute([2, 1])(repeated_word_attention)\n",
    "sentence_representation = Multiply()([bilstm, repeated_word_attention])\n",
    "sentence_representation = Lambda(lambda x: K.sum(x, axis=1))(sentence_representation)\n",
    "\n",
    "bilstm_output = Dense(1)(sentence_representation)\n",
    "\n",
    "model = Model(inputs=[input_data],\n",
    "            outputs=[bilstm_output])\n",
    "# 3. 모델 학습과정 설정하기\n",
    "model.compile(loss='mean_squared_error', optimizer='adam')\n",
    "\n",
    "hist = model.fit(x_train, y_train, epochs=10, batch_size=32, validation_data=(x_val, y_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_4 (InputLayer)            (None, 40, 1)        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "lstm_4 (LSTM)                   (None, 40, 100)      40800       input_4[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "attention_layer_4 (AttentionLay (None, 40)           10200       lstm_4[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "repeat_vector_4 (RepeatVector)  (None, 100, 40)      0           attention_layer_4[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "permute_4 (Permute)             (None, 40, 100)      0           repeat_vector_4[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "multiply_4 (Multiply)           (None, 40, 100)      0           lstm_4[0][0]                     \n",
      "                                                                 permute_4[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "lambda_3 (Lambda)               (None, 100)          0           multiply_4[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "dense_3 (Dense)                 (None, 1)            101         lambda_3[0][0]                   \n",
      "==================================================================================================\n",
      "Total params: 51,101\n",
      "Trainable params: 51,101\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train Score:  3.050595018736969e-05\n",
      "Validataion Score:  3.0229520780267195e-05\n",
      "Test Score:  3.0229520780267195e-05\n"
     ]
    }
   ],
   "source": [
    "# 5. 학습과정 살펴보기\n",
    "plt.plot(hist.history['loss'])\n",
    "plt.plot(hist.history['val_loss'])\n",
    "plt.ylim(0.0, 0.15)\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()\n",
    "\n",
    "# 6. 모델 평가하기\n",
    "trainScore = model.evaluate(x_train, y_train, verbose=0)\n",
    "model.reset_states()\n",
    "print('Train Score: ', trainScore)\n",
    "valScore = model.evaluate(x_val, y_val, verbose=0)\n",
    "model.reset_states()\n",
    "print('Validataion Score: ', valScore)\n",
    "testScore = model.evaluate(x_test, y_test, verbose=0)\n",
    "model.reset_states()\n",
    "print('Test Score: ', testScore)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 7. 모델 사용하기\n",
    "look_ahead = 250\n",
    "xhat = x_test[0]\n",
    "predictions = np.zeros((look_ahead,1))\n",
    "for i in range(look_ahead):\n",
    "    prediction = model.predict(np.array([xhat]), batch_size=1)\n",
    "    predictions[i] = prediction\n",
    "    xhat = np.vstack([xhat[1:],prediction])\n",
    "    \n",
    "plt.figure(figsize=(12,5))\n",
    "plt.plot(np.arange(look_ahead),predictions,'r',label=\"prediction\")\n",
    "plt.plot(np.arange(look_ahead),y_test[:look_ahead],label=\"test function\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Attention visualize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Attention check\n",
    "# 위의 attention layer 부분을 output으로\n",
    "attention_extractor = Model(inputs=[input_data],\n",
    "                             outputs=[attention_layer])  \n",
    "# (N, 40)\n",
    "attention_expamle = attention_extractor.predict(x_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1.2864340e-06 2.3297710e-06 3.6501749e-06 5.2106298e-06 7.0482333e-06\n",
      " 9.2856208e-06 1.2130141e-05 1.5883727e-05 2.0971740e-05 2.7994298e-05\n",
      " 3.7804832e-05 5.1625346e-05 7.1212780e-05 9.9099823e-05 1.3893853e-04\n",
      " 1.9599240e-04 2.7782962e-04 3.9529527e-04 5.6384178e-04 8.0532866e-04\n",
      " 1.1503776e-03 1.6413693e-03 2.3360685e-03 3.3117542e-03 4.6695205e-03\n",
      " 6.5380782e-03 9.0760719e-03 1.2471545e-02 1.6937118e-02 2.2699282e-02\n",
      " 2.9981662e-02 3.8981959e-02 4.9845852e-02 6.2640250e-02 7.7332310e-02\n",
      " 9.3777388e-02 1.1171997e-01 1.3080847e-01 1.5062439e-01 1.7071383e-01]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<BarContainer object of 40 artists>"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(attention_expamle[0])\n",
    "\n",
    "plt.bar(np.arange(attention_expamle.shape[1]), attention_expamle[2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}