4.TEXT_CLASSIFICATION/4.1.7.CNN_Classification.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 카글 텍스트 분류 - 합성곱 신경망 활용 접근방법"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "import os\n",
    "import numpy as np\n",
    "import json\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "import pandas as pd\n",
    "\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
    "from tensorflow.keras.preprocessing.text import Tokenizer\n",
    "from tensorflow.keras.preprocessing import sequence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_IN_PATH = './data_in/'\n",
    "DATA_OUT_PATH = './data_out/'\n",
    "TRAIN_INPUT_DATA = 'train_input.npy'\n",
    "TRAIN_LABEL_DATA = 'train_label.npy'\n",
    "TEST_INPUT_DATA = 'test_input.npy'\n",
    "TEST_ID_DATA = 'test_id.npy'\n",
    "\n",
    "DATA_CONFIGS = 'data_configs.json'\n",
    "\n",
    "train_input_data = np.load(open(DATA_IN_PATH + TRAIN_INPUT_DATA, 'rb'))\n",
    "train_label_data = np.load(open(DATA_IN_PATH + TRAIN_LABEL_DATA, 'rb'))\n",
    "test_input_data = np.load(open(DATA_IN_PATH + TEST_INPUT_DATA, 'rb'))\n",
    "\n",
    "with open(DATA_IN_PATH + DATA_CONFIGS, 'r') as f:\n",
    "    prepro_configs = json.load(f)\n",
    "    print(prepro_configs.keys())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 파라메터 변수\n",
    "RNG_SEED = 1234\n",
    "BATCH_SIZE = 16\n",
    "NUM_EPOCHS = 3\n",
    "VOCAB_SIZE = prepro_configs['vocab_size'] + 1\n",
    "EMB_SIZE = 128\n",
    "VALID_SPLIT = 0.2\n",
    "\n",
    "train_input, eval_input, train_label, eval_label = train_test_split(train_input_data, train_label_data, test_size=VALID_SPLIT, random_state=RNG_SEED)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## tf.data 세팅"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def mapping_fn(X, Y=None):\n",
    "    input, label = {'x': X}, Y\n",
    "    return input, label\n",
    "\n",
    "def train_input_fn():\n",
    "    dataset = tf.data.Dataset.from_tensor_slices((train_input, train_label))\n",
    "    dataset = dataset.shuffle(buffer_size=len(train_input))\n",
    "    dataset = dataset.batch(BATCH_SIZE)\n",
    "    dataset = dataset.map(mapping_fn)\n",
    "    dataset = dataset.repeat(count=NUM_EPOCHS)\n",
    "\n",
    "    iterator = dataset.make_one_shot_iterator()\n",
    "    \n",
    "    return iterator.get_next()\n",
    "\n",
    "def eval_input_fn():\n",
    "    dataset = tf.data.Dataset.from_tensor_slices((eval_input, eval_label))\n",
    "    dataset = dataset.shuffle(buffer_size=len(eval_input))\n",
    "    dataset = dataset.batch(BATCH_SIZE)\n",
    "    dataset = dataset.map(mapping_fn)\n",
    "\n",
    "    iterator = dataset.make_one_shot_iterator()\n",
    "    \n",
    "    return iterator.get_next()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 모델 세팅"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 모델에 대한 메인 부분입니다.\n",
    "\n",
    "\n",
    "def model_fn(features, labels, mode):\n",
    "\n",
    "    TRAIN = mode == tf.estimator.ModeKeys.TRAIN\n",
    "    EVAL = mode == tf.estimator.ModeKeys.EVAL\n",
    "    PREDICT = mode == tf.estimator.ModeKeys.PREDICT\n",
    "    \n",
    "    #embedding layer를 선언합니다.\n",
    "    embedding_layer = keras.layers.Embedding(\n",
    "                    VOCAB_SIZE,\n",
    "                    EMB_SIZE)(features['x'])\n",
    "    \n",
    "    # embedding layer에 대한 output에 대해 dropout을 취합니다.\n",
    "    dropout_emb = keras.layers.Dropout(rate=0.5)(embedding_layer)\n",
    "\n",
    "    ## filters = 128이고 kernel_size = 3,4,5입니다.\n",
    "    ## 길이가 3,4,5인 128개의 다른 필터를 생성합니다. 3,4,5 gram의 효과처럼 다양한 각도에서 문장을 보는 효과가 있습니다.\n",
    "    ## conv1d는 (배치사이즈, 길이, 채널)로 입력값을 받는데, 배치사이즈: 문장 숫자 | 길이: 각 문장의 단어의 개수 | 채널: 임베딩 출력 차원수임\n",
    "    \n",
    "    conv1 = keras.layers.Conv1D(\n",
    "         filters=128,\n",
    "         kernel_size=3,\n",
    "        padding='valid',\n",
    "         activation=tf.nn.relu)(dropout_emb)\n",
    "    \n",
    "    pool1 = keras.layers.GlobalMaxPool1D()(conv1)\n",
    "\n",
    "    conv2 = keras.layers.Conv1D(\n",
    "         filters=128,\n",
    "         kernel_size=4,\n",
    "        padding='valid',\n",
    "         activation=tf.nn.relu)(dropout_emb)\n",
    "    \n",
    "    pool2 = keras.layers.GlobalMaxPool1D()(conv2)\n",
    "    \n",
    "    conv3 = keras.layers.Conv1D(\n",
    "         filters=128,\n",
    "         kernel_size=5,\n",
    "        padding='valid',\n",
    "         activation=tf.nn.relu)(dropout_emb)\n",
    "    pool3 = keras.layers.GlobalMaxPool1D()(conv3)\n",
    "    \n",
    "    concat = keras.layers.concatenate([pool1, pool2, pool3]) #3,4,5gram이후 모아주기\n",
    "    \n",
    "    hidden = keras.layers.Dense(250, activation=tf.nn.relu)(concat)\n",
    "    dropout_hidden = keras.layers.Dropout(rate=0.5)(hidden)\n",
    "    logits = keras.layers.Dense(1, name='logits')(dropout_hidden)\n",
    "    logits = tf.squeeze(logits, axis=-1)\n",
    "    \n",
    "    #최종적으로 학습, 평가, 테스트의 단계로 나누어 활용\n",
    "    \n",
    "    if PREDICT:\n",
    "        return tf.estimator.EstimatorSpec(\n",
    "            mode=mode,\n",
    "            predictions={\n",
    "                'prob': tf.nn.sigmoid(logits)\n",
    "            }\n",
    "        )\n",
    "        \n",
    "    loss = tf.losses.sigmoid_cross_entropy(labels, logits)\n",
    "\n",
    "    if EVAL:\n",
    "        pred = tf.nn.sigmoid(logits)\n",
    "        accuracy = tf.metrics.accuracy(labels, tf.round(pred))\n",
    "        return tf.estimator.EstimatorSpec(mode=mode, loss=loss, eval_metric_ops={'acc': accuracy})\n",
    "        \n",
    "    if TRAIN:\n",
    "        global_step = tf.train.get_global_step()\n",
    "        train_op = tf.train.AdamOptimizer(0.001).minimize(loss, global_step)\n",
    "\n",
    "        return tf.estimator.EstimatorSpec(mode=mode, train_op=train_op, loss = loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_dir = os.path.join(os.getcwd(), \"data_out/checkpoint/cnn/\")\n",
    "os.makedirs(model_dir, exist_ok=True)\n",
    "\n",
    "config_tf = tf.estimator.RunConfig(save_checkpoints_steps=200, keep_checkpoint_max=2,\n",
    "                                    log_step_count_steps=400)\n",
    "\n",
    " #에스티메이터 객체 생성\n",
    "cnn_est = tf.estimator.Estimator(model_fn, model_dir=model_dir, config=config_tf)\n",
    "cnn_est.train(train_input_fn) #학습하기\n",
    "cnn_est.evaluate(eval_input_fn) #평가하기"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "test_input_data = np.load(open(DATA_IN_PATH + TEST_INPUT_DATA, 'rb')) \n",
    "ids = np.load(open(DATA_IN_PATH + TEST_ID_DATA, 'rb'))\n",
    "\n",
    "predict_input_fn = tf.estimator.inputs.numpy_input_fn(x={\"x\":test_input_data}, shuffle=False)\n",
    "\n",
    "predictions = np.array([p['prob'] for p in cnn_est.predict(input_fn=predict_input_fn)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "output = pd.DataFrame( data={\"id\": ids, \"sentiment\": predictions} )\n",
    "\n",
    "output.to_csv( DATA_OUT_PATH + \"Bag_of_Words_model_test.csv\", index=False, quoting=3 )"
   ]
  }
 ],
 "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}