text-classification-nn.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "parallel-investing",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/polinab/.virtualenvs/deep/lib/python3.8/site-packages/pandas/compat/__init__.py:97: UserWarning: Could not import the lzma module. Your installed Python is incomplete. Attempting to use lzma compression will result in a RuntimeError.\n",
      "  warnings.warn(msg)\n"
     ]
    }
   ],
   "source": [
    "#importing libraries\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import os\n",
    "from varname import nameof\n",
    "import string\n",
    "import nltk\n",
    "from nltk.corpus import stopwords\n",
    "\n",
    "#deep learning library\n",
    "from keras.models import *\n",
    "from keras.layers import *\n",
    "from keras.callbacks import *\n",
    "from keras import optimizers\n",
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras_preprocessing.sequence import pad_sequences\n",
    "from tensorflow.keras.preprocessing import sequence\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "from keras.callbacks import CSVLogger\n",
    "from tensorflow.python.keras.layers import Conv1D, Conv2D, MaxPooling1D, Flatten\n",
    "\n",
    "seed = 1337"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "twenty-munich",
   "metadata": {},
   "source": [
    "### Import and explore data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "verified-remains",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "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>text</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Wow... Loved this place.</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Crust is not good.</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Not tasty and the texture was just nasty.</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Stopped by during the late May bank holiday of...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>The selection on the menu was great and so wer...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                text  label\n",
       "0                           Wow... Loved this place.      1\n",
       "1                                 Crust is not good.      0\n",
       "2          Not tasty and the texture was just nasty.      0\n",
       "3  Stopped by during the late May bank holiday of...      1\n",
       "4  The selection on the menu was great and so wer...      1"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "yelp_labelled = pd.read_csv('yelp_labelled.txt', sep='\\t', header=None)\n",
    "yelp_labelled.columns = ['text', 'label']\n",
    "yelp_labelled.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "combined-speech",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:xlabel='label', ylabel='Count'>"
      ]
     },
     "execution_count": 3,
     "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": [
    "sns.histplot(data=yelp_labelled, x=\"label\",bins=2,binwidth=0.4)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "appropriate-eight",
   "metadata": {},
   "source": [
    "- 0 and 1 labels are equally distributed, so we do not need to resample the data\n",
    "- 0 is a bad review, 1 is a good review\n",
    "- small set of data, so we can use all of it"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "recovered-produce",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "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>text</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>226</th>\n",
       "      <td>This is an unbelievable BARGAIN!</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>129</th>\n",
       "      <td>Definitely worth venturing off the strip for t...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>So they performed.</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>847</th>\n",
       "      <td>In fact I'm going to round up to 4 stars, just...</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>554</th>\n",
       "      <td>However, my recent experience at this particul...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                  text  label\n",
       "226                   This is an unbelievable BARGAIN!      1\n",
       "129  Definitely worth venturing off the strip for t...      1\n",
       "24                                  So they performed.      1\n",
       "847  In fact I'm going to round up to 4 stars, just...      1\n",
       "554  However, my recent experience at this particul...      0"
      ]
     },
     "execution_count": 4,
     "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": [
    "y = yelp_labelled['label']\n",
    "train_df, test_df = train_test_split(yelp_labelled, test_size=0.2)\n",
    "sns.histplot(data=yelp_labelled, x=\"label\",bins=2,binwidth=0.4)\n",
    "train_df.head()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "advance-creature",
   "metadata": {},
   "source": [
    "### Vectorize the word data: building my own word dictionary\n",
    "- a neural network model cannot interpret words, they need to be encoded into numbers\n",
    "- external libraries exist that do this, like Word2Vec\n",
    "- After playing around with the libraries I decided to make my own encoding to fully understand how the basics work\n",
    "- getting rid of stop words made the model perform worse. Understandable since sequentially many of the stop words actually add information to the phrases! "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "alert-instrumentation",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "149"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# separate into positive and negative posts\n",
    "positive_posts = pd.Series.to_numpy(yelp_labelled[yelp_labelled.label == 1]['text'])\n",
    "negative_posts = pd.Series.to_numpy(yelp_labelled[yelp_labelled.label == 0]['text'])\n",
    "\n",
    "# keep an object with both sets for later use\n",
    "posts = np.append(negative_posts, positive_posts, axis=0)\n",
    "\n",
    "longest_post = max(len(post) for post in posts)\n",
    "longest_post\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "satisfactory-thomas",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "def unique_words(lines):\n",
    "    '''\n",
    "    returns dictionary of unique words in input\n",
    "    assigns an integer key to each word\n",
    "    does not include punctuation or stop words \n",
    "    stop words can be removed or included depending on if it benefits the model\n",
    "    '''\n",
    "    unique_words = set()\n",
    "    stop_words = set(stopwords.words('english'))\n",
    "\n",
    "    for line in lines:\n",
    "        for word in line:\n",
    "#             if word not in stop_words:\n",
    "#                 unique_words.add(word)\n",
    "            unique_words.add(word)\n",
    "    word_vals = set(i for i in range(len(unique_words)))\n",
    "    word_keys = dict(zip(unique_words, word_vals))\n",
    "    return word_keys\n",
    "\n",
    "def word_to_vec(posts):\n",
    "    '''\n",
    "    encodes word input into integer arrays \n",
    "    stop words can be removed or included depending on the model\n",
    "    '''\n",
    "    stop_words = set(stopwords.words('english'))\n",
    "    posts = [line.lower().translate(str.maketrans('', '', string.punctuation)).split(' ') for line in posts]\n",
    "    word_dict = unique_words(posts)\n",
    "    word_key_map = [[word_dict[word] for word in post] for post in posts]\n",
    "#     word_key_map = [[word_dict[word] for word in post if word not in stop_words] for post in posts]\n",
    "    return word_key_map"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "generous-drill",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\"Crust is not good.\" becomes [88, 469, 1347, 20]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "<__array_function__ internals>:5: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray\n"
     ]
    }
   ],
   "source": [
    "neg_encoded = word_to_vec(negative_posts)\n",
    "pos_encoded = word_to_vec(positive_posts)\n",
    "word_vecs = np.concatenate((neg_encoded, pos_encoded))\n",
    "print(f'\"{negative_posts[0]}\" becomes {word_vecs[0]}')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "creative-police",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "# Padding the data samples to a maximum review length in words\n",
    "longest_post = max([len(post) for post in word_to_vec(posts)])\n",
    "padded_word_vecs = sequence.pad_sequences(word_vecs, maxlen=longest_post)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "electrical-luxembourg",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train shape: (670, 32) (670,)\n",
      "X_test shape: (330, 32) (330,)\n"
     ]
    }
   ],
   "source": [
    "# re-create labels, concatenate with data, and set up test/train split\n",
    "labels = np.concatenate((np.zeros(len(neg_encoded)), np.ones(len(pos_encoded))))\n",
    "X_train, X_test, y_train, y_test = train_test_split(padded_word_vecs,labels,test_size=0.33)\n",
    "\n",
    "print('X_train shape:', X_train.shape, y_train.shape)\n",
    "print('X_test shape:', X_test.shape, y_test.shape)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "lightweight-jordan",
   "metadata": {},
   "source": [
    "#### Modularize model fitting and display functions\n",
    "- I set up the following functions to avoid code redundancy and simplify the model evaluation process\n",
    "- the functions allow for fitting, accuracy evaluation, and plotting of accuracy and loss over epochs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "alive-incentive",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "def fit_model(model, e, v, x_train, y_train, x_test, y_test):\n",
    "    '''\n",
    "    fits model over epochs and stores results in csv file\n",
    "    '''\n",
    "    csv_logger = CSVLogger(f\"{nameof(model)}.csv\", append=False)\n",
    "\n",
    "    model.fit(\n",
    "        x_train,\n",
    "        y_train, \n",
    "        epochs=e, \n",
    "                   verbose=v,\n",
    "                   callbacks=[csv_logger],\n",
    "                   validation_data=(\n",
    "                       x_test,\n",
    "                       y_test))\n",
    "\n",
    "def model_acc(model,x,y):\n",
    "    scores = model.evaluate(x,y)\n",
    "    return(scores[1])\n",
    "    \n",
    "def acc_graphs(model):\n",
    "    '''\n",
    "    displays accuracy and loss of model fit over epochs\n",
    "    '''\n",
    "    df = pd.read_csv(f\"{nameof(model)}.csv\")\n",
    "    df.set_index('epoch', inplace=True)\n",
    "    \n",
    "    fig, ax =plt.subplots(1,2, figsize=(10, 5))\n",
    "    df[['accuracy','val_accuracy']].plot(ax = ax[0])\n",
    "    df[['loss','val_loss']].plot(ax = ax[1])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "subtle-moderator",
   "metadata": {},
   "source": [
    "### Word-wise LSTM Model with custom embedding\n",
    "LSTM is an artificial RNN architecture.\n",
    "reference: https://towardsdatascience.com/a-beginners-guide-on-sentiment-analysis-with-rnn-9e100627c02e"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "charitable-configuration",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding (Embedding)        (None, 32, 32)            66560     \n",
      "_________________________________________________________________\n",
      "lstm (LSTM)                  (None, 32)                8320      \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 1)                 33        \n",
      "=================================================================\n",
      "Total params: 74,913\n",
      "Trainable params: 74,913\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "11/11 [==============================] - 0s 3ms/step - loss: 0.0361 - accuracy: 0.9879\n",
      "Test accuracy: 0.9878787994384766\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "vocab_size = max([max(arr) for arr in word_to_vec(posts)])\n",
    "embedding_size = 32\n",
    "num_epochs = 20\n",
    "lstm_model = Sequential()\n",
    "lstm_model.add(Embedding(vocab_size, embedding_size, input_length=longest_post))\n",
    "lstm_model.add(LSTM(embedding_size))\n",
    "lstm_model.add(Dropout(0.5))\n",
    "lstm_model.add(Dense(1, activation='sigmoid'))\n",
    "print(lstm_model.summary())\n",
    "\n",
    "lstm_model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['accuracy'])\n",
    "\n",
    "csv_logger = CSVLogger(\"lstm_model.csv\", append=False)\n",
    "\n",
    "fit_model(lstm_model, num_epochs,0,X_train, y_train,X_test,y_test)\n",
    "\n",
    "print('Test accuracy:', model_acc(lstm_model,X_test, y_test))\n",
    "\n",
    "acc_graphs(lstm_model)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "secret-insured",
   "metadata": {},
   "source": [
    "### Word-wise CNN Model with custom embedding\n",
    "https://machinelearningmastery.com/develop-word-embedding-model-predicting-movie-review-sentiment/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "freelance-palace",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_1 (Embedding)      (None, 32, 50)            104000    \n",
      "_________________________________________________________________\n",
      "conv1d (Conv1D)              (None, 25, 50)            20050     \n",
      "_________________________________________________________________\n",
      "max_pooling1d (MaxPooling1D) (None, 12, 50)            0         \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 600)               0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 10)                6010      \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 1)                 11        \n",
      "=================================================================\n",
      "Total params: 130,071\n",
      "Trainable params: 130,071\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "11/11 [==============================] - 0s 3ms/step - loss: 0.0677 - accuracy: 0.9788\n",
      "Test accuracy: 0.978787899017334\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "embedding_size = 50\n",
    "cnn_model = Sequential()\n",
    "cnn_model.add(Embedding(vocab_size, embedding_size, input_length=longest_post))\n",
    "cnn_model.add(Conv1D(embedding_size,kernel_size=8, activation='relu'))\n",
    "cnn_model.add(MaxPooling1D(pool_size=2))\n",
    "cnn_model.add(Flatten())\n",
    "cnn_model.add(Dense(10, activation='relu'))\n",
    "cnn_model.add(Dense(1, activation='sigmoid'))\n",
    "cnn_model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['accuracy'])\n",
    "print(cnn_model.summary())\n",
    "\n",
    "fit_model(cnn_model, num_epochs,0,X_train, y_train,X_test,y_test)\n",
    "\n",
    "print('Test accuracy:', model_acc(cnn_model,X_test, y_test))\n",
    "\n",
    "acc_graphs(cnn_model)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "veterinary-efficiency",
   "metadata": {},
   "source": [
    "##### Problems with this approach:\n",
    "- after coding this section, I realized that using the entire set to build the embedding dictionary is \"cheating\": in the real world, we would not have access to the test data\n",
    "- I decided to see how much the results would be affected by separating train and test data before building the dictionary\n",
    "- unfortunately, the results were much worse. This makes sense because missing words would have no assocation to a label\n",
    "- this is why we want to use similarity vectors for words.\n",
    "- as you can see below, the in-sample accuracy converges to nearly 1, while the validation accuracy hovers around 50%\n",
    "- below, you can see how my custom-dictionary model fails when the dictionary is built only with training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "cognitive-assembly",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "<__array_function__ internals>:5: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray\n",
      "<__array_function__ internals>:5: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train shape: (800, 149) (800,)\n",
      "X_test shape: (200, 149) (200,)\n",
      "Model: \"sequential_2\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_2 (Embedding)      (None, 32, 32)            66560     \n",
      "_________________________________________________________________\n",
      "lstm_1 (LSTM)                (None, 32)                8320      \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 1)                 33        \n",
      "=================================================================\n",
      "Total params: 74,913\n",
      "Trainable params: 74,913\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 32) for input KerasTensor(type_spec=TensorSpec(shape=(None, 32), dtype=tf.float32, name='embedding_2_input'), name='embedding_2_input', description=\"created by layer 'embedding_2_input'\"), but it was called on an input with incompatible shape (32, 149).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 32) for input KerasTensor(type_spec=TensorSpec(shape=(None, 32), dtype=tf.float32, name='embedding_2_input'), name='embedding_2_input', description=\"created by layer 'embedding_2_input'\"), but it was called on an input with incompatible shape (32, 149).\n",
      "WARNING:tensorflow:Model was constructed with shape (None, 32) for input KerasTensor(type_spec=TensorSpec(shape=(None, 32), dtype=tf.float32, name='embedding_2_input'), name='embedding_2_input', description=\"created by layer 'embedding_2_input'\"), but it was called on an input with incompatible shape (None, 149).\n",
      "7/7 [==============================] - 0s 13ms/step - loss: 1.3844 - accuracy: 0.5600\n",
      "Test accuracy: 0.5600000023841858\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def preprocess(df):\n",
    "    # separate into positive and negative posts\n",
    "    positive_posts = pd.Series.to_numpy(df[df.label == 1]['text'])\n",
    "    negative_posts = pd.Series.to_numpy(df[df.label == 0]['text'])\n",
    "\n",
    "    # keep an object with both sets for later use\n",
    "    posts = np.append(negative_posts, positive_posts, axis=0)\n",
    "    \n",
    "    neg_encoded = word_to_vec(negative_posts)\n",
    "    pos_encoded = word_to_vec(positive_posts)\n",
    "    word_vecs = np.concatenate((neg_encoded, pos_encoded))\n",
    "    \n",
    "    # Padding the data samples to a maximum review length in words\n",
    "#     longest_post = max([len(post) for post in word_to_vec(posts)])\n",
    "    padded_word_vecs = sequence.pad_sequences(word_vecs, maxlen=149)\n",
    "    \n",
    "    labels = np.concatenate((np.zeros(len(neg_encoded)), np.ones(len(pos_encoded))))\n",
    "    \n",
    "    return padded_word_vecs,labels\n",
    "\n",
    "X_train, y_train = preprocess(train_df)\n",
    "X_test, y_test = preprocess(test_df)\n",
    "\n",
    "print('X_train shape:', X_train.shape, y_train.shape)\n",
    "print('X_test shape:', X_test.shape, y_test.shape)\n",
    "\n",
    "vocab_size = max([max(arr) for arr in word_to_vec(posts)])\n",
    "embedding_size = 32\n",
    "lstm_model = Sequential()\n",
    "lstm_model.add(Embedding(vocab_size, embedding_size, input_length=longest_post))\n",
    "lstm_model.add(LSTM(embedding_size))\n",
    "lstm_model.add(Dropout(0.5))\n",
    "lstm_model.add(Dense(1, activation='sigmoid'))\n",
    "print(lstm_model.summary())\n",
    "\n",
    "lstm_model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['accuracy'])\n",
    "\n",
    "fit_model(lstm_model, num_epochs,0,X_train, y_train,X_test,y_test)\n",
    "\n",
    "print('Test accuracy:', model_acc(lstm_model,X_test, y_test))\n",
    "\n",
    "acc_graphs(lstm_model)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "provincial-musical",
   "metadata": {},
   "source": [
    "### Create word-wise GloVe pre-trained embeddings\n",
    "Re-doing the character-wise model with GloVe pre-trained embeddings, since the custom dictionary approach fails when applied on training set only."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "spoken-sacramento",
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "class wordVec:\n",
    "    '''\n",
    "    generates a split of training and test data in class format.\n",
    "    useful when dealing with many sets of data in different splits \n",
    "    '''\n",
    "    def __init__(self, data):\n",
    "        self.samples = pd.Series.to_numpy(data.text)\n",
    "        self.labels = pd.Series.to_numpy(data.label)\n",
    "        \n",
    "        self.train_samples = None\n",
    "        self.val_samples = None\n",
    "        \n",
    "        self.train_labels = None\n",
    "        self.val_labels = None\n",
    "\n",
    "    def split(self, n):\n",
    "        rng = np.random.RandomState(seed)\n",
    "        rng.shuffle(self.samples)\n",
    "        rng = np.random.RandomState(seed)\n",
    "        rng.shuffle(self.labels)\n",
    "\n",
    "        # Extract a training & validation split\n",
    "        validation_split = n\n",
    "        num_validation_samples = int(validation_split * len(self.samples))\n",
    "        self.train_samples = self.samples[:-num_validation_samples]\n",
    "        self.val_samples = self.samples[-num_validation_samples:]\n",
    "        self.train_labels = self.labels[:-num_validation_samples]\n",
    "        self.val_labels = self.labels[-num_validation_samples:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "architectural-northwest",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'Overpriced for what you are getting.'"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Split to train and test\n",
    "\n",
    "yelp_labelled = pd.read_csv('yelp_labelled.txt', sep='\\t', header=None)\n",
    "yelp_labelled.columns = ['text', 'label']\n",
    "\n",
    "yelpVec = wordVec(yelp_labelled)\n",
    "yelpVec.split(0.2)\n",
    "yelpVec.train_samples[0]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "superb-shopper",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1816\n"
     ]
    }
   ],
   "source": [
    "#Tokenize the sentences\n",
    "tokenizer = Tokenizer()\n",
    "\n",
    "#preparing vocabulary\n",
    "tokenizer.fit_on_texts(list(yelpVec.train_samples))\n",
    "\n",
    "#converting text into integer sequences\n",
    "x_tr_seq  = tokenizer.texts_to_sequences(yelpVec.train_samples) \n",
    "x_val_seq = tokenizer.texts_to_sequences(yelpVec.val_samples) \n",
    "\n",
    "#padding to prepare sequences of same length\n",
    "x_tr_seq  = pad_sequences(x_tr_seq, maxlen=100)\n",
    "x_val_seq = pad_sequences(x_val_seq, maxlen=100)\n",
    "\n",
    "size_of_vocabulary=len(tokenizer.word_index) + 1 #+1 for padding\n",
    "print(size_of_vocabulary)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "mature-language",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train = np.array(x_tr_seq),\n",
    "y_train = np.array(yelpVec.train_labels)\n",
    "\n",
    "X_test = np.array(x_val_seq),\n",
    "y_test = np.array(yelpVec.val_labels)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "ignored-wichita",
   "metadata": {},
   "outputs": [],
   "source": [
    "def load_pretrained():\n",
    "    path_to_glove_file = os.path.join(\n",
    "        os.path.expanduser(\"~\"), \".keras/datasets/glove.6B.300d.txt\")\n",
    "\n",
    "    embeddings_index = {}\n",
    "    with open(path_to_glove_file) as f:\n",
    "        for line in f:\n",
    "            word, coefs = line.split(maxsplit=1)\n",
    "            coefs = np.fromstring(coefs, \"f\", sep=\" \")\n",
    "            embeddings_index[word] = coefs\n",
    "\n",
    "    print(\"Found %s word vectors.\" % len(embeddings_index))\n",
    "    return embeddings_index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "second-reporter",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 400000 word vectors.\n"
     ]
    }
   ],
   "source": [
    "embed_index = load_pretrained()\n",
    "\n",
    "# create a weight matrix for words in training docs\n",
    "embedding_matrix = np.zeros((size_of_vocabulary, 300))\n",
    "\n",
    "for word, i in tokenizer.word_index.items():\n",
    "    embedding_vector = embed_index.get(word)\n",
    "    if embedding_vector is not None:\n",
    "        embedding_matrix[i] = embedding_vector\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "surprised-missile",
   "metadata": {},
   "source": [
    "### Word-Wise LSTM model using GloVe embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "standing-clock",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_3\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_3 (Embedding)      (None, 100, 300)          544800    \n",
      "_________________________________________________________________\n",
      "lstm_2 (LSTM)                (None, 100, 128)          219648    \n",
      "_________________________________________________________________\n",
      "global_max_pooling1d (Global (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 1)                 65        \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 772,769\n",
      "Trainable params: 227,969\n",
      "Non-trainable params: 544,800\n",
      "_________________________________________________________________\n",
      "None\n",
      "7/7 [==============================] - 0s 36ms/step - loss: 0.5272 - accuracy: 0.8450\n",
      "Test accuracy: 0.8450000286102295\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "embedding_size = 50\n",
    "lstm_glove_model = Sequential()\n",
    "lstm_glove_model.add(Embedding(size_of_vocabulary,300,weights=[embedding_matrix],input_length=100,trainable=False))\n",
    "lstm_glove_model.add(LSTM(128,return_sequences=True,dropout=0.2))\n",
    "lstm_glove_model.add(GlobalMaxPooling1D())\n",
    "lstm_glove_model.add(Dense(64,activation='relu')) \n",
    "lstm_glove_model.add(Dense(1,activation='sigmoid')) \n",
    "lstm_glove_model.add(Flatten())\n",
    "lstm_glove_model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['accuracy'])\n",
    "print(lstm_glove_model.summary())\n",
    "\n",
    "num_epochs = 20\n",
    "fit_model(lstm_glove_model, num_epochs,0,X_train, y_train,X_test,y_test)\n",
    "\n",
    "print('Test accuracy:', model_acc(lstm_glove_model,X_test, y_test))\n",
    "\n",
    "acc_graphs(lstm_glove_model)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "accessible-immunology",
   "metadata": {},
   "source": [
    "### Word-wise CNN using GloVe embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "defined-classroom",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_4\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_4 (Embedding)      (None, 100, 300)          544800    \n",
      "_________________________________________________________________\n",
      "conv1d_1 (Conv1D)            (None, 93, 50)            120050    \n",
      "_________________________________________________________________\n",
      "max_pooling1d_1 (MaxPooling1 (None, 46, 50)            0         \n",
      "_________________________________________________________________\n",
      "flatten_2 (Flatten)          (None, 2300)              0         \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 50)                115050    \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 10)                510       \n",
      "_________________________________________________________________\n",
      "dense_8 (Dense)              (None, 5)                 55        \n",
      "_________________________________________________________________\n",
      "dense_9 (Dense)              (None, 1)                 6         \n",
      "=================================================================\n",
      "Total params: 780,471\n",
      "Trainable params: 235,671\n",
      "Non-trainable params: 544,800\n",
      "_________________________________________________________________\n",
      "None\n",
      "7/7 [==============================] - 0s 10ms/step - loss: 0.5801 - accuracy: 0.7600\n",
      "Test accuracy: 0.7599999904632568\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "embedding_size = 50\n",
    "cnn_glove_model = Sequential()\n",
    "cnn_glove_model.add(Embedding(size_of_vocabulary,300,weights=[embedding_matrix],input_length=100,trainable=False))\n",
    "cnn_glove_model.add(Conv1D(embedding_size,kernel_size=8, activation='relu'))\n",
    "cnn_glove_model.add(MaxPooling1D(pool_size=2))\n",
    "cnn_glove_model.add(Flatten())\n",
    "cnn_glove_model.add(Dense(50, activation='sigmoid'))\n",
    "cnn_glove_model.add(Dense(10, activation='relu'))\n",
    "cnn_glove_model.add(Dense(5, activation='sigmoid'))\n",
    "cnn_glove_model.add(Dense(1, activation='sigmoid'))\n",
    "cnn_glove_model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['accuracy'])\n",
    "print(cnn_glove_model.summary())\n",
    "\n",
    "fit_model(cnn_glove_model, num_epochs,0,X_train, y_train,X_test,y_test)\n",
    "\n",
    "print('Test accuracy:', model_acc(cnn_glove_model,X_test, y_test))\n",
    "\n",
    "acc_graphs(cnn_glove_model)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fluid-vulnerability",
   "metadata": {},
   "source": [
    "### Prepare and vectorize the character-wise data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "virtual-civilization",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "def char_tokenizer(lines):\n",
    "    tk = Tokenizer(num_words=None, char_level=True, oov_token='UNK')\n",
    "    lines_low = [[l.lower() for l in lines]]\n",
    "    tk.fit_on_texts(lines_low)\n",
    "\n",
    "    alphabet = \"abcdefghijklmnopqrstuvwxyz0123456789,;.!?:'\\\"/\\\\|_@#$%^&*~`+-=<>()[]{}\"\n",
    "    char_dict = {}\n",
    "    for i, char in enumerate(alphabet):\n",
    "        char_dict[char] = i + 1\n",
    "\n",
    "    # Use char_dict to replace the tk.word_index\n",
    "    tk.word_index = char_dict.copy()\n",
    "    # Add 'UNK' to the vocabulary\n",
    "    tk.word_index[tk.oov_token] = max(char_dict.values()) + 1\n",
    "    return tk"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "dense-youth",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "def embed_weights(tk):\n",
    "    vocab_size = len(tk.word_index)\n",
    "    embedding_weights = []\n",
    "    embedding_weights.append(np.zeros(vocab_size))\n",
    "\n",
    "    for char,i in tk.word_index.items():\n",
    "        onehot = np.zeros(vocab_size)\n",
    "        onehot[i-1] = 1\n",
    "        embedding_weights.append(onehot)\n",
    "\n",
    "    return np.array(embedding_weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "acute-scale",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wow... Loved this place. becomes [23, 15, 23, 39, 39, 39, 69, 12, 15, 22, 5, 4, 69, 20, 8, 9, 19, 69, 16, 12, 1, 3, 5, 39]\n"
     ]
    }
   ],
   "source": [
    "# transform reviews into character encodings:\n",
    "\n",
    "posts = positive_posts + negative_posts\n",
    "tk = char_tokenizer(posts)\n",
    "vocab_size = len(tk.word_index)\n",
    "pos_seqs = tk.texts_to_sequences(positive_posts)\n",
    "neg_seqs = tk.texts_to_sequences(negative_posts)\n",
    "print(f'{positive_posts[0]} becomes {pos_seqs[0]}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "preceding-optimum",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "longest post is 149 characters long\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/polinab/.virtualenvs/deep/lib/python3.8/site-packages/numpy/core/_asarray.py:136: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray\n",
      "  return array(a, dtype, copy=False, order=order, subok=True)\n",
      "<__array_function__ internals>:5: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray\n"
     ]
    }
   ],
   "source": [
    "longest_post = max([len(x) for x in np.append(neg_seqs, pos_seqs, axis=0)])\n",
    "print(f'longest post is {longest_post} characters long')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "prescription-wrapping",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([23., 15., 23., 39., 39., 39., 69., 12., 15., 22.,  5.,  4., 69.,\n",
       "       20.,  8.,  9., 19., 69., 16., 12.,  1.,  3.,  5., 39.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "        0.,  0.,  0.,  0.,  0.,  0.], dtype=float32)"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# pad data:\n",
    "\n",
    "pos_seq_padded = pad_sequences(pos_seqs, maxlen=longest_post, padding='post')\n",
    "neg_seq_padded = pad_sequences(neg_seqs, maxlen=longest_post, padding='post')\n",
    "\n",
    "pos_data = np.array(pos_seq_padded, dtype='float32')\n",
    "neg_data = np.array(neg_seq_padded, dtype='float32')\n",
    "pos_data[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "eleven-shower",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train_char shape: (670, 149) (670,)\n",
      "X_test_char shape: (330, 149) (330,)\n"
     ]
    }
   ],
   "source": [
    "padded_char_vecs = np.append(neg_data, pos_data, axis = 0)\n",
    "\n",
    "labels = np.concatenate((np.zeros(len(neg_data)), np.ones(len(pos_data))))\n",
    "X_train_char, X_test_char, y_train_char, y_test_char = train_test_split(padded_char_vecs, labels, test_size=0.33)\n",
    "\n",
    "print('X_train_char shape:', X_train_char.shape, y_train_char.shape)\n",
    "print('X_test_char shape:', X_test_char.shape, y_test_char.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "indie-pride",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "# define embedding layer \n",
    "def embed_weights(tk):\n",
    "    vocab_size = len(tk.word_index)\n",
    "    embedding_weights = []\n",
    "    embedding_weights.append(np.zeros(vocab_size))\n",
    "\n",
    "    for char,i in tk.word_index.items():\n",
    "        onehot = np.zeros(vocab_size)\n",
    "        onehot[i-1] = 1\n",
    "        embedding_weights.append(onehot)\n",
    "\n",
    "    return np.array(embedding_weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "formal-serbia",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "embedding_size = 69 #TODO: find where this number comes from\n",
    "input_size = longest_post\n",
    "em_weights = embed_weights(tk)\n",
    "embedding_layer = Embedding(vocab_size+1,embedding_size,input_length=input_size,weights=[em_weights])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "separated-release",
   "metadata": {},
   "source": [
    "###  Character-wise CNN model\n",
    "https://realpython.com/python-keras-text-classification/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "proved-cassette",
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_5\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "embedding_5 (Embedding)      (None, 149, 69)           4830      \n",
      "_________________________________________________________________\n",
      "conv1d_2 (Conv1D)            (None, 142, 69)           38157     \n",
      "_________________________________________________________________\n",
      "global_max_pooling1d_1 (Glob (None, 69)                0         \n",
      "_________________________________________________________________\n",
      "flatten_3 (Flatten)          (None, 69)                0         \n",
      "_________________________________________________________________\n",
      "dense_10 (Dense)             (None, 10)                700       \n",
      "_________________________________________________________________\n",
      "dense_11 (Dense)             (None, 1)                 11        \n",
      "=================================================================\n",
      "Total params: 43,698\n",
      "Trainable params: 43,698\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "11/11 [==============================] - 0s 5ms/step - loss: 0.6009 - accuracy: 0.7515\n",
      "Test accuracy: 0.7515151500701904\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cnn_char_model = Sequential()\n",
    "cnn_char_model.add(embedding_layer)\n",
    "cnn_char_model.add(Conv1D(embedding_size, kernel_size=8, activation='relu'))\n",
    "cnn_char_model.add(GlobalMaxPooling1D())\n",
    "\n",
    "cnn_char_model.add(Flatten())\n",
    "cnn_char_model.add(Dense(10, activation='relu'))\n",
    "cnn_char_model.add(Dense(1, activation='sigmoid'))\n",
    "cnn_char_model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])\n",
    "print(cnn_char_model.summary())\n",
    "\n",
    "fit_model(cnn_char_model, num_epochs,0,\n",
    "          X_train_char, y_train_char,\n",
    "          X_test_char,y_test_char)\n",
    "\n",
    "print('Test accuracy:', model_acc(cnn_char_model,X_test_char, y_test_char))\n",
    "\n",
    "acc_graphs(cnn_char_model)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "absent-kitchen",
   "metadata": {},
   "source": [
    "### Conclusions\n",
    "\n",
    "#### Embeddings\n",
    "Out of the four models I tried, LSTM using GloVe embeddings led to the best results. I did have even better results with LSTM using a custom word dictionary, but I had to discard those results after I realized that training the dictionary on the full set was not appropriate to a real-life use case. In a real machine learning scenario, I would not have access to the test data and I would not be able to construct a dictionary using information from the test set. Since this data set was small, and ever review contained different vocabulary, not including the test set in the training dictionary led to a lot of unmatched words. If the set of data contained less variability among words between train and test sets, the results of the custom dictionary could have been improved. <br>\n",
    "I found very good results using the GloVe pre-trained word embeddings. I used the 300-dimensional ones, which I believe allowed for more information to be used by the model that a lower dimension. I also tried word2vec embeddings chose to go with GloVe in the end due to ease of use. Since GloVe uses a pre-existing large body of words, it does not depend as much on the information in the input data.\n",
    "\n",
    "#### Choice of model\n",
    "When comparing the word vector approach, LSTMs performed better than CNNs. Furthermore, I found that an LSTM model performed better than a regular RNN, but the regular RNN also performed about as well as a CNN. LSTMs retain more imformation over time than RNNs through the use of additional \"memory cell\" units (Chung et al 2014), which allow better improvements over multiple epochs. <br>RNNs perform better than CNNs when the input data contains longer, semantically complicated sentences (Yin et al 2017). This makes sense because RNNs are usually used for predicting sequences, while CNNs were originally used on images and explore spatial correlation of data points. When a sentence needs to be parsed semantically, I would expect an RNN to perform better. However, if detection of specific features is more important, then I would choose a CNN. \n",
    "\n",
    "\n",
    "#### Hyperparameters and layers\n",
    "While I did not do a full hyperparameter grid search integration, I did explore a few combinations of layers and hyperparameters when building my models. Out of what I explored, here is what I found the biggest improvements with:\n",
    "- max pooling layer : dimensionality reduction. This had the greatest effect on increasing accuracy.\n",
    "- adding more dense layers with numbers of neurons between 5 and 50 : models with 2-3 layers performed slightly better than single-layer\n",
    "- combining a relu with a sigmoid layer in the stack, instead of only using one kind or the other. I also tried softmax and tanh activation functions, but they did not yield better results.\n",
    "- hidden size and batch size adjustments, but no clear trend on which works better.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "social-chamber",
   "metadata": {},
   "source": [
    "#### Future improvements\n",
    "\n",
    "If I had more time, I would focus on looking into the input layer of the model and the dimensions of the data I am passing it. <br>\n",
    "If I were to continue this project, I would also build a hyperparameter grid search for my successful models. This way, I would be able to automate the process of tuning hyperparameters instead of trying different numbers manually. Ideally, I would find a way to explore different numbers of layers as well. <br>\n",
    "Finally, I would explore stacking a convolutional layer with an LSTM layer. When I was exploring different architectures I found many references to combined models with CNN and LSTM (Friedrich et al 2020), and I would expect good results in this use case."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "wanted-mechanism",
   "metadata": {},
   "source": [
    "#### Reference Papers\n",
    "- http://www.shl-dataset.org/wp-content/uploads/SHLChallenge2020/short_videos_presentations/01_ubicomp20q-sub1022-cam-i7.pdf\n",
    "- https://arxiv.org/pdf/1702.01923.pdf\n",
    "- https://arxiv.org/abs/1412.3555"
   ]
  }
 ],
 "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}