BiRNN_LSTM_GRU-Experiments.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "T28QEDT-Uqj2"
   },
   "source": [
    "<p align=\"center\">\n",
    " <img src=\"http://www.di.uoa.gr/themes/corporate_lite/logo_el.png\" title=\"Department of Informatics and Telecommunications - University of Athens\"/> </p>\n",
    "\n",
    "---\n",
    "<h1 align=\"center\"> \n",
    "  Artificial Intelligence\n",
    "</h1>\n",
    "<h1 align=\"center\" > \n",
    "  Deep Learning for Natural Language Processing\n",
    "</h1>\n",
    "\n",
    "---\n",
    "<h2 align=\"center\"> \n",
    " <b>Konstantinos Nikoletos</b>\n",
    "</h2>\n",
    "\n",
    "<h3 align=\"center\"> \n",
    " <b>Winter 2020-2021</b>\n",
    "</h3>\n",
    "\n",
    "\n",
    "---\n",
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Wk9SOEk6mCXZ"
   },
   "source": [
    "\n",
    "#  <center> <b>  Experiment notebook | GloVe with RNN  </b> </center>\n",
    "\n",
    "### __Task__ \n",
    "> Develop a sentiment classifier using a bidirectional stacked RNN with LSTM/GRU cells\n",
    "for the Twitter sentiment analysis dataset of the previous assignment available here. For\n",
    "the development of the models, you can experiment with the number of stacked RNNs,\n",
    "the number of hidden layers, type of cells, skip connections, gradient clipping and dropout\n",
    "probability. Document the performance of different configurations on your final report.\n",
    "Use the Adam optimizer and the binary cross-entropy loss function. Remember to transform the predicted logits to probabilities using a sigmoid function. You should also utilize\n",
    "pre-trained word embeddings (GloVe) as the initial embeddings to input on your models.\n",
    "\n",
    "For the best model you found:\n",
    " - Compute precision, recall and F1 for each class.\n",
    " - Plot the loss vs. epochs curve and the ROC curve and explain what you see.\n",
    " - Compare with your best model from Homework 2.\n",
    " \n",
    "Your solution should be implemented in PyTorch and we expect your report to be well\n",
    "documented.\n",
    "\n",
    "### __Notebook__ \n",
    "In this notebook I am going to run the multiple combinations of models and parameters.\n",
    "\n",
    "---\n",
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "6OfAccyufasd"
   },
   "source": [
    "__Import__ of essential libraries\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "dVSHtKTrfGiV",
    "outputId": "fde33d1e-2d04-4adc-cdf9-4ed338dbf699"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
      "[nltk_data]   Unzipping tokenizers/punkt.zip.\n",
      "[nltk_data] Downloading package wordnet to /root/nltk_data...\n",
      "[nltk_data]   Unzipping corpora/wordnet.zip.\n",
      "[nltk_data] Downloading package stopwords to /root/nltk_data...\n",
      "[nltk_data]   Unzipping corpora/stopwords.zip.\n",
      "[nltk_data] Downloading package averaged_perceptron_tagger to\n",
      "[nltk_data]     /root/nltk_data...\n",
      "[nltk_data]   Unzipping taggers/averaged_perceptron_tagger.zip.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 1,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd \n",
    "import sys # only needed to determine Python version number\n",
    "import matplotlib # only needed to determine Matplotlib version \n",
    "import nltk\n",
    "from nltk.stem import WordNetLemmatizer\n",
    "nltk.download('punkt')\n",
    "nltk.download('wordnet')\n",
    "nltk.download('stopwords')\n",
    "nltk.download('averaged_perceptron_tagger')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "PF7UeY--wfi2"
   },
   "source": [
    "__PyTorch imports__\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "RX9EjAJ_wekG"
   },
   "outputs": [],
   "source": [
    "import pprint\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "from torchtext import data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "vC9DWskzConr"
   },
   "source": [
    "Selecting device (GPU - CUDA if available)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "NiRkpLhiCntl",
    "outputId": "6d96bfa6-0bc9-4026-e1f5-424bccc3a1c9"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training on GPU.\n"
     ]
    }
   ],
   "source": [
    "# First checking if GPU is available\n",
    "train_on_gpu=torch.cuda.is_available()\n",
    "\n",
    "if(train_on_gpu):\n",
    "    print('Training on GPU.')\n",
    "else:\n",
    "    print('No GPU available, training on CPU.')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "jAyyovNsGMh7"
   },
   "source": [
    "Initializing random seed as in lesson's notebooks.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "57CtIpBzF-9i"
   },
   "outputs": [],
   "source": [
    "import random\n",
    "import os\n",
    "\n",
    "def set_seed(seed = 1234):\n",
    "    '''Sets the seed of the entire notebook so results are the same every time we run.\n",
    "    This is for REPRODUCIBILITY.'''\n",
    "    np.random.seed(seed)\n",
    "    random.seed(seed)\n",
    "    torch.manual_seed(seed)\n",
    "    torch.cuda.manual_seed(seed)\n",
    "    # When running on the CuDNN backend, two further options must be set\n",
    "    torch.backends.cudnn.deterministic = True\n",
    "    # Set a fixed value for the hash seed\n",
    "    os.environ['PYTHONHASHSEED'] = str(seed)\n",
    "    \n",
    "set_seed()\n",
    "device = 'cuda'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "FHPD7L2off6x"
   },
   "source": [
    "# __Loading__ the dataset\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "CsY-8ThJfmIa",
    "outputId": "da5fa256-d6e5-46c1-cdf1-1a423092dcc5"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mounted at /content/drive\n"
     ]
    }
   ],
   "source": [
    "# Opening data file\n",
    "import io\n",
    "from google.colab import drive\n",
    "drive.mount('/content/drive',force_remount=True)\n",
    "filePath = r\"/content/drive/My Drive/Colab_Notebooks/SentimentTweets.csv\"\n",
    "csv = pd.read_csv(filePath,nrows=50000)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "sn3gutNLh_84"
   },
   "source": [
    "From the whole dataset I use 50.000 tweets because of the large time of execution needed from the whole dataset (1.200.000+).\n",
    "\n",
    "I mostly used GPU, but some days I had multiple problems with runtime and Colab for reasons I didn't understand. For this reason I executed the notebook with small sizes in order to see it all run. !Warning! CPU runtime takes a lot."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "YkMXhFN6fl4H"
   },
   "source": [
    "This dataset encoded the target variable with a 3-point ordinal scale: 0 = negative, 2 = neutral, 4 = positive. But this dataset contains only:\n",
    "- 0 -> negative\n",
    "- 4 -> positive\n",
    "\n",
    "So, tweets will be predicted only as positive or negative\n",
    "\n",
    "\n",
    " Forming __dataframe__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 204
    },
    "id": "_oid1uQAf-xZ",
    "outputId": "35746a08-fd16-404c-a1b4-8786785e4ba4"
   },
   "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>target</th>\n",
       "      <th>text</th>\n",
       "      <th>tweet</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>#brokenpromises...</td>\n",
       "      <td>#brokenpromises...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>David Carradine  so sad. Thai's law not sure i...</td>\n",
       "      <td>David Carradine  so sad. Thai's law not sure i...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4</td>\n",
       "      <td>A @ 415 B @ 425. Tell your bro i say congrats!</td>\n",
       "      <td>A @ 415 B @ 425. Tell your bro i say congrats!</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>@littlefluffycat  Indeed.</td>\n",
       "      <td>@littlefluffycat  Indeed.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>Completed Race 4 Life in 58mins with girlies f...</td>\n",
       "      <td>Completed Race 4 Life in 58mins with girlies f...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   target  ...                                              tweet\n",
       "0       0  ...                                #brokenpromises... \n",
       "1       0  ...  David Carradine  so sad. Thai's law not sure i...\n",
       "2       4  ...    A @ 415 B @ 425. Tell your bro i say congrats! \n",
       "3       4  ...                          @littlefluffycat  Indeed.\n",
       "4       4  ...  Completed Race 4 Life in 58mins with girlies f...\n",
       "\n",
       "[5 rows x 3 columns]"
      ]
     },
     "execution_count": 6,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tweetDf  = pd.DataFrame(data=csv)\n",
    "tweetDf.drop(columns=[\"Unnamed: 0\",\"id\",\"date\",\"flag\",\"user\"],inplace=True)\n",
    "tweetDf['tweet'] = tweetDf['text']\n",
    "tweetDf.head(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "vc56d1wI5lMz"
   },
   "source": [
    "#  Data pre-processing\n",
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Qv-s7sFSm7_F"
   },
   "source": [
    "### __Text cleaning__ from stopwords,links and pancutation.\n",
    "\n",
    "Creating a set of stopwords first consisted of nltk library and wordcloud library:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "ZCAjk7u0m9gC"
   },
   "outputs": [],
   "source": [
    "from wordcloud import STOPWORDS as stopwords1,WordCloud\n",
    "from nltk.corpus import stopwords as stopwords2\n",
    "from nltk import word_tokenize\n",
    "import string\n",
    "import unicodedata\n",
    "import re\n",
    "import string\n",
    "            \n",
    "            \n",
    "stopwords = set(stopwords1)\n",
    "stopwords_nltk =  set(stopwords2.words('english'))\n",
    "stopwords.update(stopwords_nltk)\n",
    "\n",
    "# Removes stopwords from sentences\n",
    "def removeStopWords(text):\n",
    "    return ' '.join([y for y in word_tokenize(text) if y not in stopwords])\n",
    "\n",
    "# Function that lemmatizes wanted columns\n",
    "lemmatizer = nltk.stem.WordNetLemmatizer()\n",
    "def Lemmatize(text):\n",
    "    return ' '.join([lemmatizer.lemmatize(y,'v') for y in word_tokenize(text)])\n",
    "\n",
    "def normalazation(text):\n",
    "    return ' '.join([normalize_word(y) for y in word_tokenize(text)])\n",
    "\n",
    "def normalize_word(word):\n",
    "    # Will remove the encode token by token\n",
    "    word = unicodedata.normalize('NFKD', word).encode('ASCII','ignore').decode('ASCII')\n",
    "    return word\n",
    "\n",
    "def smallwordsRemoval(text):\n",
    "    return ' '.join([remove_smallWords(y) for y in word_tokenize(text)])\n",
    "\n",
    "def remove_smallWords(word):\n",
    "    if (len(word) > 2):\n",
    "        return word\n",
    "    else:\n",
    "        return ''    \n",
    "\n",
    "def preprocess_text(column_of_sentences):\n",
    "    lemmatizer = nltk.stem.WordNetLemmatizer()\n",
    "    processed_column = [] \n",
    "    for sentence in column_of_sentences:\n",
    "      processed_sentence = [] \n",
    "      for word in word_tokenize(sentence):\n",
    "        if word not in string.punctuation:\n",
    "          word = word.replace(r'@\\S+',' ').replace(r'\\\\\\S+',' ').replace('\\n',' ').replace('[{}]'.format(string.punctuation), '')\n",
    "          word = word.lower()\n",
    "          word = ''.join(ch for ch in word if not ch in string.punctuation and not ch.isdigit())\n",
    "          word = unicodedata.normalize('NFKD', word).encode('ASCII','ignore').decode('ASCII')\n",
    "          word = lemmatizer.lemmatize(word,'v')\n",
    "          if len(word) > 2:\n",
    "            processed_sentence.append(word)\n",
    "      if len(processed_sentence) == 0:\n",
    "        processed_column.append(np.nan)\n",
    "      else:\n",
    "        processed_column.append(' '.join(processed_sentence))\n",
    "    return processed_column        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 204
    },
    "id": "_ydr09vXo_hM",
    "outputId": "5ce8807e-b5c2-493f-df89-9b4285bb37bc"
   },
   "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>target</th>\n",
       "      <th>text</th>\n",
       "      <th>tweet</th>\n",
       "      <th>processed_tweets</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>#brokenpromises...</td>\n",
       "      <td>#brokenpromises...</td>\n",
       "      <td>brokenpromises</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>David Carradine  so sad. Thai's law not sure i...</td>\n",
       "      <td>David Carradine  so sad. Thai's law not sure i...</td>\n",
       "      <td>david carradine sad thai law not sure fowl pla...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4</td>\n",
       "      <td>A @ 415 B @ 425. Tell your bro i say congrats!</td>\n",
       "      <td>A @ 415 B @ 425. Tell your bro i say congrats!</td>\n",
       "      <td>tell your bro say congrats</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>@littlefluffycat  Indeed.</td>\n",
       "      <td>@littlefluffycat  Indeed.</td>\n",
       "      <td>littlefluffycat indeed</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>Completed Race 4 Life in 58mins with girlies f...</td>\n",
       "      <td>Completed Race 4 Life in 58mins with girlies f...</td>\n",
       "      <td>complete race life mins with girlies from work...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   target  ...                                   processed_tweets\n",
       "0       0  ...                                     brokenpromises\n",
       "1       0  ...  david carradine sad thai law not sure fowl pla...\n",
       "2       4  ...                         tell your bro say congrats\n",
       "3       4  ...                             littlefluffycat indeed\n",
       "4       4  ...  complete race life mins with girlies from work...\n",
       "\n",
       "[5 rows x 4 columns]"
      ]
     },
     "execution_count": 8,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tweetDf['processed_tweets'] = preprocess_text(tweetDf['tweet'])\n",
    "\n",
    "# Dropping empty lines\n",
    "tweetDf.dropna(subset=['processed_tweets'],inplace=True)\n",
    "\n",
    "tweetDf.head(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "FyWfWF1EcEcE"
   },
   "source": [
    "Checking average and maximum number of words in tweets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "bIBRSk2XbwBQ",
    "outputId": "b1a0e21a-7470-4d4b-93a5-d494305046e6"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Zero-length reviews: 0\n",
      "Maximum review length: 139\n",
      "Average review length:  69\n"
     ]
    }
   ],
   "source": [
    "from statistics import mean\n",
    "from collections import Counter\n",
    "\n",
    "# Outlier review stats\n",
    "tweet_lens = Counter([len(x) for x in tweetDf['processed_tweets']])\n",
    "print(\"Zero-length reviews: {}\".format(tweet_lens[0]))\n",
    "print(\"Maximum review length: {}\".format(max(tweet_lens)))\n",
    "print(\"Average review length: \",int(mean([x for x,y in tweet_lens.items()])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "WZiGdTU4kzVl"
   },
   "source": [
    "### WordCloud\n",
    "--- \n",
    "For a better evaluation of data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 303
    },
    "id": "aBzor5pZkzEO",
    "outputId": "9e23aa8a-cc12-4134-f3e7-ed28ddca05fd"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import collections\n",
    "# Function that given a list of data will produce a WordCloud\n",
    "def generate_WordCloud(data,stopwords):\n",
    "    textWords = []\n",
    "    for x in data:\n",
    "        for y in word_tokenize(x) :\n",
    "            y = y.lower()\n",
    "            y = WordNetLemmatizer().lemmatize(y)\n",
    "            textWords.append(y)\n",
    "                \n",
    "    counter = collections.Counter(textWords)\n",
    "    words = WordCloud(background_color ='white',stopwords=stopwords,margin=0).generate_from_frequencies(frequencies=dict(counter))\n",
    "\n",
    "    plt.figure(figsize=(10,5))\n",
    "    plt.imshow(words,interpolation='bilinear')\n",
    "    plt.axis(\"off\")\n",
    "    plt.show()\n",
    "generate_WordCloud(tweetDf['processed_tweets'],stopwords)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "pDtbIy5UnALM"
   },
   "source": [
    "### __Label encoder__ for target **values**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "Ze983LVwnArD"
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import LabelEncoder\n",
    "le = LabelEncoder() \n",
    "le.fit(tweetDf['target'])    # change string to number for categories\n",
    "tweetDf['target'] = le.transform(tweetDf['target'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "l6Xtu8Ztv_y0"
   },
   "source": [
    "Encoding\n",
    "- 0 : 0 -> Negative \n",
    "- 1 : 4 -> Positive "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 204
    },
    "id": "u1FXGwHx1Bo8",
    "outputId": "d21e4530-09c2-42d2-f089-e7c0cbf07caa"
   },
   "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>target</th>\n",
       "      <th>text</th>\n",
       "      <th>tweet</th>\n",
       "      <th>processed_tweets</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>#brokenpromises...</td>\n",
       "      <td>#brokenpromises...</td>\n",
       "      <td>brokenpromises</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>David Carradine  so sad. Thai's law not sure i...</td>\n",
       "      <td>David Carradine  so sad. Thai's law not sure i...</td>\n",
       "      <td>david carradine sad thai law not sure fowl pla...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>A @ 415 B @ 425. Tell your bro i say congrats!</td>\n",
       "      <td>A @ 415 B @ 425. Tell your bro i say congrats!</td>\n",
       "      <td>tell your bro say congrats</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>@littlefluffycat  Indeed.</td>\n",
       "      <td>@littlefluffycat  Indeed.</td>\n",
       "      <td>littlefluffycat indeed</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1</td>\n",
       "      <td>Completed Race 4 Life in 58mins with girlies f...</td>\n",
       "      <td>Completed Race 4 Life in 58mins with girlies f...</td>\n",
       "      <td>complete race life mins with girlies from work...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   target  ...                                   processed_tweets\n",
       "0       0  ...                                     brokenpromises\n",
       "1       0  ...  david carradine sad thai law not sure fowl pla...\n",
       "2       1  ...                         tell your bro say congrats\n",
       "3       1  ...                             littlefluffycat indeed\n",
       "4       1  ...  complete race life mins with girlies from work...\n",
       "\n",
       "[5 rows x 4 columns]"
      ]
     },
     "execution_count": 12,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tweetDf.head(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "9wp8npXJgBSh"
   },
   "source": [
    "# __Encoding__ the words\n",
    "\n",
    "In this model I will use GloVe"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "TSbQcZmJCwRo"
   },
   "source": [
    "Import of libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "RYdjLy1OCd95"
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import numpy as np\n",
    "from sklearn.feature_extraction.text import CountVectorizer,TfidfVectorizer\n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "from torch.utils.data import TensorDataset,DataLoader\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import accuracy_score,f1_score,recall_score,precision_score\n",
    "from torchtext import vocab\n",
    "from torchtext.data import Field,Dataset,LabelField,BucketIterator,TabularDataset # TorchText has Text processing Function"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "oFG5JJUkgzwU"
   },
   "source": [
    "#### Loading GloVe embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "-swIbwKvwTla"
   },
   "outputs": [],
   "source": [
    "from torchtext.vocab import GloVe"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "1Nnx8VKewT5X"
   },
   "outputs": [],
   "source": [
    "# glove_25 = GloVe(name='twitter.27B', dim=25)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "mXXr158caayl"
   },
   "outputs": [],
   "source": [
    "# glove_50 = GloVe(name='twitter.27B', dim=50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "bG9zZQWswJ_J"
   },
   "outputs": [],
   "source": [
    "# glove_100 = GloVe(name='twitter.27B', dim=100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "yJg7OVOawOba"
   },
   "outputs": [],
   "source": [
    "# glove_200 = GloVe(name='twitter.27B', dim=200)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "hTKXadT1J04M",
    "outputId": "c720ab9d-6172-44c0-8bf3-abdfc7c1523f"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|█████████▉| 1190757/1193514 [00:40<00:00, 44025.10it/s]\n",
      "  0%|          | 0/1193514 [00:00<?, ?it/s]\u001b[A\n",
      "  0%|          | 2553/1193514 [00:00<00:46, 25520.41it/s]\u001b[A\n",
      "  0%|          | 4649/1193514 [00:00<00:49, 23954.24it/s]\u001b[A\n",
      "  1%|          | 6781/1193514 [00:00<00:51, 23096.21it/s]\u001b[A\n",
      "  1%|          | 9240/1193514 [00:00<00:50, 23522.27it/s]\u001b[A\n",
      "  1%|          | 11220/1193514 [00:00<00:53, 22263.48it/s]\u001b[A\n",
      "  1%|          | 13480/1193514 [00:00<00:52, 22362.82it/s]\u001b[A\n",
      "  1%|▏         | 15554/1193514 [00:00<00:53, 21848.64it/s]\u001b[A\n",
      "  1%|▏         | 17544/1193514 [00:00<00:56, 20691.86it/s]\u001b[A\n",
      "  2%|▏         | 19975/1193514 [00:00<00:54, 21658.15it/s]\u001b[A\n",
      "  2%|▏         | 22222/1193514 [00:01<00:53, 21893.36it/s]\u001b[A\n",
      "  2%|▏         | 24448/1193514 [00:01<00:53, 22000.89it/s]\u001b[A\n",
      "  2%|▏         | 26798/1193514 [00:01<00:52, 22430.12it/s]\u001b[A\n",
      "  2%|▏         | 29290/1193514 [00:01<00:50, 23122.99it/s]\u001b[A\n",
      "  3%|▎         | 31750/1193514 [00:01<00:49, 23546.93it/s]\u001b[A\n",
      "  3%|▎         | 34099/1193514 [00:01<00:52, 22295.15it/s]\u001b[A\n",
      "  3%|▎         | 36472/1193514 [00:01<00:50, 22704.85it/s]\u001b[A\n",
      "  3%|▎         | 38755/1193514 [00:01<00:50, 22741.04it/s]\u001b[A\n",
      "  3%|▎         | 41238/1193514 [00:01<00:49, 23328.24it/s]\u001b[A\n",
      "  4%|▎         | 43579/1193514 [00:01<00:49, 23197.13it/s]\u001b[A\n",
      "  4%|▍         | 45905/1193514 [00:02<00:50, 22912.23it/s]\u001b[A\n",
      "  4%|▍         | 48202/1193514 [00:02<00:51, 22438.05it/s]\u001b[A\n",
      "  4%|▍         | 50452/1193514 [00:02<00:50, 22425.44it/s]\u001b[A\n",
      "  4%|▍         | 52880/1193514 [00:02<00:49, 22947.10it/s]\u001b[A\n",
      "  5%|▍         | 55181/1193514 [00:02<00:49, 22892.53it/s]\u001b[A\n",
      "  5%|▍         | 57641/1193514 [00:02<00:48, 23379.22it/s]\u001b[A\n",
      "  5%|▌         | 59985/1193514 [00:02<00:49, 22779.55it/s]\u001b[A\n",
      "  5%|▌         | 62271/1193514 [00:02<00:49, 22719.13it/s]\u001b[A\n",
      "  5%|▌         | 64548/1193514 [00:02<00:52, 21517.33it/s]\u001b[A\n",
      "  6%|▌         | 66846/1193514 [00:02<00:51, 21934.40it/s]\u001b[A\n",
      "  6%|▌         | 69281/1193514 [00:03<00:49, 22606.15it/s]\u001b[A\n",
      "  6%|▌         | 71779/1193514 [00:03<00:48, 23267.14it/s]\u001b[A\n",
      "  6%|▌         | 74121/1193514 [00:03<00:49, 22822.64it/s]\u001b[A\n",
      "  6%|▋         | 76416/1193514 [00:03<00:51, 21729.46it/s]\u001b[A\n",
      "  7%|▋         | 78844/1193514 [00:03<00:49, 22436.47it/s]\u001b[A\n",
      "  7%|▋         | 81163/1193514 [00:03<00:49, 22656.22it/s]\u001b[A\n",
      "  7%|▋         | 83619/1193514 [00:03<00:47, 23195.41it/s]\u001b[A\n",
      "  7%|▋         | 86167/1193514 [00:03<00:46, 23834.66it/s]\u001b[A\n",
      "  7%|▋         | 88712/1193514 [00:03<00:45, 24295.38it/s]\u001b[A\n",
      "  8%|▊         | 91154/1193514 [00:04<00:46, 23733.12it/s]\u001b[A\n",
      "  8%|▊         | 93606/1193514 [00:04<00:45, 23963.17it/s]\u001b[A\n",
      "  8%|▊         | 96106/1193514 [00:04<00:45, 24263.02it/s]\u001b[A\n",
      "  8%|▊         | 98539/1193514 [00:04<00:49, 21970.65it/s]\u001b[A\n",
      "  8%|▊         | 100783/1193514 [00:04<00:49, 21994.50it/s]\u001b[A\n",
      "  9%|▊         | 103015/1193514 [00:04<00:50, 21446.69it/s]\u001b[A\n",
      "  9%|▉         | 105385/1193514 [00:04<00:49, 22076.05it/s]\u001b[A\n",
      "  9%|▉         | 107847/1193514 [00:04<00:47, 22781.64it/s]\u001b[A\n",
      "  9%|▉         | 110147/1193514 [00:04<00:48, 22382.62it/s]\u001b[A\n",
      "  9%|▉         | 112402/1193514 [00:04<00:48, 22424.50it/s]\u001b[A\n",
      " 10%|▉         | 114699/1193514 [00:05<00:47, 22583.15it/s]\u001b[A\n",
      " 10%|▉         | 117094/1193514 [00:05<00:46, 22975.16it/s]\u001b[A\n",
      " 10%|█         | 119439/1193514 [00:05<00:46, 23111.71it/s]\u001b[A\n",
      " 10%|█         | 121834/1193514 [00:05<00:45, 23356.35it/s]\u001b[A\n",
      " 10%|█         | 124286/1193514 [00:05<00:45, 23693.08it/s]\u001b[A\n",
      " 11%|█         | 126663/1193514 [00:05<00:44, 23715.57it/s]\u001b[A\n",
      " 11%|█         | 129038/1193514 [00:05<00:46, 23053.86it/s]\u001b[A\n",
      " 11%|█         | 131350/1193514 [00:05<00:48, 22069.08it/s]\u001b[A\n",
      " 11%|█         | 133774/1193514 [00:05<00:46, 22677.78it/s]\u001b[A\n",
      " 11%|█▏        | 136095/1193514 [00:05<00:46, 22832.70it/s]\u001b[A\n",
      " 12%|█▏        | 138439/1193514 [00:06<00:45, 23010.92it/s]\u001b[A\n",
      " 12%|█▏        | 140748/1193514 [00:06<00:47, 22137.07it/s]\u001b[A\n",
      " 12%|█▏        | 143277/1193514 [00:06<00:45, 22995.53it/s]\u001b[A\n",
      " 12%|█▏        | 145594/1193514 [00:06<00:45, 23024.84it/s]\u001b[A\n",
      " 12%|█▏        | 147909/1193514 [00:06<00:50, 20895.55it/s]\u001b[A\n",
      " 13%|█▎        | 150045/1193514 [00:06<00:49, 21002.92it/s]\u001b[A\n",
      " 13%|█▎        | 152492/1193514 [00:06<00:47, 21934.94it/s]\u001b[A\n",
      " 13%|█▎        | 154719/1193514 [00:06<00:49, 21153.06it/s]\u001b[A\n",
      " 13%|█▎        | 156864/1193514 [00:06<00:52, 19862.70it/s]\u001b[A\n",
      " 13%|█▎        | 159181/1193514 [00:07<00:49, 20750.18it/s]\u001b[A\n",
      " 14%|█▎        | 161545/1193514 [00:07<00:47, 21538.80it/s]\u001b[A\n",
      " 14%|█▎        | 163865/1193514 [00:07<00:46, 22010.70it/s]\u001b[A\n",
      " 14%|█▍        | 166196/1193514 [00:07<00:45, 22380.56it/s]\u001b[A\n",
      " 14%|█▍        | 168454/1193514 [00:07<00:46, 22117.17it/s]\u001b[A\n",
      " 14%|█▍        | 170773/1193514 [00:07<00:45, 22427.57it/s]\u001b[A\n",
      " 14%|█▍        | 173027/1193514 [00:07<00:45, 22242.85it/s]\u001b[A\n",
      " 15%|█▍        | 175416/1193514 [00:07<00:44, 22712.54it/s]\u001b[A\n",
      " 15%|█▍        | 177696/1193514 [00:07<00:46, 21790.02it/s]\u001b[A\n",
      " 15%|█▌        | 179889/1193514 [00:08<00:47, 21222.32it/s]\u001b[A\n",
      " 15%|█▌        | 182216/1193514 [00:08<00:46, 21796.47it/s]\u001b[A\n",
      " 15%|█▌        | 184730/1193514 [00:08<00:44, 22702.04it/s]\u001b[A\n",
      " 16%|█▌        | 187196/1193514 [00:08<00:43, 23254.87it/s]\u001b[A\n",
      " 16%|█▌        | 189538/1193514 [00:08<00:43, 23195.28it/s]\u001b[A\n",
      " 16%|█▌        | 191987/1193514 [00:08<00:42, 23564.48it/s]\u001b[A\n",
      " 16%|█▋        | 194353/1193514 [00:08<00:44, 22702.91it/s]\u001b[A\n",
      " 16%|█▋        | 196787/1193514 [00:08<00:43, 23168.47it/s]\u001b[A\n",
      " 17%|█▋        | 199340/1193514 [00:08<00:41, 23828.93it/s]\u001b[A\n",
      " 17%|█▋        | 201736/1193514 [00:08<00:42, 23380.94it/s]\u001b[A\n",
      " 17%|█▋        | 204121/1193514 [00:09<00:42, 23519.42it/s]\u001b[A\n",
      " 17%|█▋        | 206481/1193514 [00:09<00:45, 21478.81it/s]\u001b[A\n",
      " 17%|█▋        | 208669/1193514 [00:09<00:45, 21485.16it/s]\u001b[A\n",
      " 18%|█▊        | 211175/1193514 [00:09<00:43, 22443.88it/s]\u001b[A\n",
      " 18%|█▊        | 213451/1193514 [00:09<00:43, 22308.09it/s]\u001b[A\n",
      " 18%|█▊        | 215704/1193514 [00:09<00:44, 21898.33it/s]\u001b[A\n",
      " 18%|█▊        | 217911/1193514 [00:09<00:47, 20370.65it/s]\u001b[A\n",
      " 18%|█▊        | 220126/1193514 [00:09<00:46, 20873.51it/s]\u001b[A\n",
      " 19%|█▊        | 222267/1193514 [00:09<00:46, 21031.25it/s]\u001b[A\n",
      " 19%|█▉        | 224389/1193514 [00:10<00:48, 20180.41it/s]\u001b[A\n",
      " 19%|█▉        | 226428/1193514 [00:10<00:52, 18342.19it/s]\u001b[A\n",
      " 19%|█▉        | 228476/1193514 [00:10<00:50, 18933.62it/s]\u001b[A\n",
      " 19%|█▉        | 230726/1193514 [00:10<00:48, 19878.10it/s]\u001b[A\n",
      " 20%|█▉        | 233109/1193514 [00:10<00:45, 20917.96it/s]\u001b[A\n",
      " 20%|█▉        | 235503/1193514 [00:10<00:44, 21740.86it/s]\u001b[A\n",
      " 20%|█▉        | 237840/1193514 [00:10<00:43, 22204.71it/s]\u001b[A\n",
      " 20%|██        | 240090/1193514 [00:10<00:44, 21377.84it/s]\u001b[A\n",
      " 20%|██        | 242255/1193514 [00:10<00:44, 21210.53it/s]\u001b[A\n",
      " 20%|██        | 244576/1193514 [00:10<00:43, 21770.76it/s]\u001b[A\n",
      " 21%|██        | 246771/1193514 [00:11<00:44, 21128.87it/s]\u001b[A\n",
      " 21%|██        | 249203/1193514 [00:11<00:42, 21993.72it/s]\u001b[A\n",
      " 21%|██        | 251584/1193514 [00:11<00:41, 22507.86it/s]\u001b[A\n",
      " 21%|██▏       | 253853/1193514 [00:11<00:41, 22560.71it/s]\u001b[A\n",
      " 21%|██▏       | 256267/1193514 [00:11<00:40, 23012.34it/s]\u001b[A\n",
      " 22%|██▏       | 258579/1193514 [00:11<00:41, 22499.96it/s]\u001b[A\n",
      " 22%|██▏       | 260839/1193514 [00:11<00:42, 22113.63it/s]\u001b[A\n",
      " 22%|██▏       | 263103/1193514 [00:11<00:41, 22267.33it/s]\u001b[A\n",
      " 22%|██▏       | 265565/1193514 [00:11<00:40, 22922.30it/s]\u001b[A\n",
      " 22%|██▏       | 268008/1193514 [00:11<00:39, 23354.60it/s]\u001b[A\n",
      " 23%|██▎       | 270352/1193514 [00:12<00:40, 22822.73it/s]\u001b[A\n",
      " 23%|██▎       | 272643/1193514 [00:12<00:41, 21960.21it/s]\u001b[A\n",
      " 23%|██▎       | 274921/1193514 [00:12<00:41, 22198.43it/s]\u001b[A\n",
      " 23%|██▎       | 277151/1193514 [00:12<00:41, 22215.65it/s]\u001b[A\n",
      " 23%|██▎       | 279509/1193514 [00:12<00:40, 22606.28it/s]\u001b[A\n",
      " 24%|██▎       | 281777/1193514 [00:12<00:41, 22039.68it/s]\u001b[A\n",
      " 24%|██▍       | 284156/1193514 [00:12<00:40, 22535.97it/s]\u001b[A\n",
      " 24%|██▍       | 286418/1193514 [00:12<00:40, 22492.66it/s]\u001b[A\n",
      " 24%|██▍       | 288875/1193514 [00:12<00:39, 23075.08it/s]\u001b[A\n",
      " 24%|██▍       | 291219/1193514 [00:13<00:38, 23182.12it/s]\u001b[A\n",
      " 25%|██▍       | 293543/1193514 [00:13<00:39, 22553.72it/s]\u001b[A\n",
      " 25%|██▍       | 295935/1193514 [00:13<00:39, 22946.70it/s]\u001b[A\n",
      " 25%|██▍       | 298237/1193514 [00:13<00:40, 22035.15it/s]\u001b[A\n",
      " 25%|██▌       | 300454/1193514 [00:13<00:41, 21703.15it/s]\u001b[A\n",
      " 25%|██▌       | 302729/1193514 [00:13<00:40, 22005.49it/s]\u001b[A\n",
      " 26%|██▌       | 304938/1193514 [00:13<00:42, 20980.61it/s]\u001b[A\n",
      " 26%|██▌       | 307339/1193514 [00:13<00:40, 21804.79it/s]\u001b[A\n",
      " 26%|██▌       | 309737/1193514 [00:13<00:39, 22413.24it/s]\u001b[A\n",
      " 26%|██▌       | 312005/1193514 [00:13<00:39, 22490.71it/s]\u001b[A\n",
      " 26%|██▋       | 314267/1193514 [00:14<00:39, 21993.77it/s]\u001b[A\n",
      " 27%|██▋       | 316478/1193514 [00:14<00:41, 20976.29it/s]\u001b[A\n",
      " 27%|██▋       | 318761/1193514 [00:14<00:40, 21498.75it/s]\u001b[A\n",
      " 27%|██▋       | 321115/1193514 [00:14<00:39, 22072.48it/s]\u001b[A\n",
      " 27%|██▋       | 323338/1193514 [00:14<00:41, 21213.94it/s]\u001b[A\n",
      " 27%|██▋       | 325517/1193514 [00:14<00:40, 21381.44it/s]\u001b[A\n",
      " 27%|██▋       | 327670/1193514 [00:14<00:40, 21424.61it/s]\u001b[A\n",
      " 28%|██▊       | 330049/1193514 [00:14<00:39, 22082.56it/s]\u001b[A\n",
      " 28%|██▊       | 332392/1193514 [00:14<00:38, 22469.16it/s]\u001b[A\n",
      " 28%|██▊       | 334649/1193514 [00:15<00:38, 22238.36it/s]\u001b[A\n",
      " 28%|██▊       | 336881/1193514 [00:15<00:40, 21281.19it/s]\u001b[A\n",
      " 28%|██▊       | 339024/1193514 [00:15<00:40, 21325.48it/s]\u001b[A\n",
      " 29%|██▊       | 341281/1193514 [00:15<00:39, 21682.36it/s]\u001b[A\n",
      " 29%|██▉       | 343458/1193514 [00:15<00:41, 20487.22it/s]\u001b[A\n",
      " 29%|██▉       | 345527/1193514 [00:15<00:41, 20295.52it/s]\u001b[A\n",
      " 29%|██▉       | 347823/1193514 [00:15<00:40, 21026.22it/s]\u001b[A\n",
      " 29%|██▉       | 350012/1193514 [00:15<00:39, 21277.64it/s]\u001b[A\n",
      " 30%|██▉       | 352153/1193514 [00:15<00:40, 20681.13it/s]\u001b[A\n",
      " 30%|██▉       | 354383/1193514 [00:15<00:39, 21139.56it/s]\u001b[A\n",
      " 30%|██▉       | 356784/1193514 [00:16<00:38, 21924.98it/s]\u001b[A\n",
      " 30%|███       | 358992/1193514 [00:16<00:38, 21870.27it/s]\u001b[A\n",
      " 30%|███       | 361334/1193514 [00:16<00:37, 22310.93it/s]\u001b[A\n",
      " 30%|███       | 363575/1193514 [00:16<00:37, 22339.21it/s]\u001b[A\n",
      " 31%|███       | 365816/1193514 [00:16<00:37, 21815.40it/s]\u001b[A\n",
      " 31%|███       | 368005/1193514 [00:16<00:38, 21176.88it/s]\u001b[A\n",
      " 31%|███       | 370132/1193514 [00:16<00:39, 20820.98it/s]\u001b[A\n",
      " 31%|███       | 372222/1193514 [00:16<00:41, 19613.61it/s]\u001b[A\n",
      " 31%|███▏      | 374509/1193514 [00:16<00:39, 20487.25it/s]\u001b[A\n",
      " 32%|███▏      | 376801/1193514 [00:17<00:38, 21159.08it/s]\u001b[A\n",
      " 32%|███▏      | 379158/1193514 [00:17<00:37, 21827.28it/s]\u001b[A\n",
      " 32%|███▏      | 381362/1193514 [00:17<00:37, 21879.33it/s]\u001b[A\n",
      " 32%|███▏      | 383565/1193514 [00:17<00:37, 21575.89it/s]\u001b[A\n",
      " 32%|███▏      | 385815/1193514 [00:17<00:36, 21842.31it/s]\u001b[A\n",
      " 33%|███▎      | 388133/1193514 [00:17<00:36, 22223.88it/s]\u001b[A\n",
      " 33%|███▎      | 390364/1193514 [00:17<00:36, 21894.09it/s]\u001b[A\n",
      " 33%|███▎      | 392674/1193514 [00:17<00:36, 22241.58it/s]\u001b[A\n",
      " 33%|███▎      | 394905/1193514 [00:17<00:37, 21350.91it/s]\u001b[A\n",
      " 33%|███▎      | 397052/1193514 [00:17<00:37, 21092.78it/s]\u001b[A\n",
      " 33%|███▎      | 399171/1193514 [00:18<00:38, 20578.10it/s]\u001b[A\n",
      " 34%|███▎      | 401238/1193514 [00:18<00:41, 19159.15it/s]\u001b[A\n",
      " 34%|███▍      | 403455/1193514 [00:18<00:39, 19972.62it/s]\u001b[A\n",
      " 34%|███▍      | 405856/1193514 [00:18<00:37, 21033.04it/s]\u001b[A\n",
      " 34%|███▍      | 408177/1193514 [00:18<00:36, 21640.57it/s]\u001b[A\n",
      " 34%|███▍      | 410370/1193514 [00:18<00:36, 21690.03it/s]\u001b[A\n",
      " 35%|███▍      | 412704/1193514 [00:18<00:35, 22157.96it/s]\u001b[A\n",
      " 35%|███▍      | 415029/1193514 [00:18<00:34, 22474.51it/s]\u001b[A\n",
      " 35%|███▍      | 417380/1193514 [00:18<00:34, 22773.19it/s]\u001b[A\n",
      " 35%|███▌      | 419668/1193514 [00:18<00:35, 21680.36it/s]\u001b[A\n",
      " 35%|███▌      | 421854/1193514 [00:19<00:35, 21613.79it/s]\u001b[A\n",
      " 36%|███▌      | 424028/1193514 [00:19<00:38, 20165.24it/s]\u001b[A\n",
      " 36%|███▌      | 426310/1193514 [00:19<00:36, 20892.31it/s]\u001b[A\n",
      " 36%|███▌      | 428426/1193514 [00:19<00:37, 20409.19it/s]\u001b[A\n",
      " 36%|███▌      | 430488/1193514 [00:19<00:37, 20079.72it/s]\u001b[A\n",
      " 36%|███▋      | 432757/1193514 [00:19<00:36, 20795.99it/s]\u001b[A\n",
      " 36%|███▋      | 434855/1193514 [00:19<00:38, 19514.43it/s]\u001b[A\n",
      " 37%|███▋      | 436898/1193514 [00:19<00:38, 19779.98it/s]\u001b[A\n",
      " 37%|███▋      | 439284/1193514 [00:19<00:36, 20848.65it/s]\u001b[A\n",
      " 37%|███▋      | 441552/1193514 [00:20<00:35, 21364.40it/s]\u001b[A\n",
      " 37%|███▋      | 443713/1193514 [00:20<00:36, 20467.73it/s]\u001b[A\n",
      " 37%|███▋      | 445785/1193514 [00:20<00:36, 20430.46it/s]\u001b[A\n",
      " 38%|███▊      | 448168/1193514 [00:20<00:34, 21341.73it/s]\u001b[A\n",
      " 38%|███▊      | 450325/1193514 [00:20<00:34, 21252.55it/s]\u001b[A\n",
      " 38%|███▊      | 452625/1193514 [00:20<00:34, 21745.86it/s]\u001b[A\n",
      " 38%|███▊      | 454814/1193514 [00:20<00:34, 21279.48it/s]\u001b[A\n",
      " 38%|███▊      | 456955/1193514 [00:20<00:34, 21201.29it/s]\u001b[A\n",
      " 38%|███▊      | 459288/1193514 [00:20<00:33, 21796.98it/s]\u001b[A\n",
      " 39%|███▊      | 461658/1193514 [00:20<00:32, 22333.44it/s]\u001b[A\n",
      " 39%|███▉      | 463902/1193514 [00:21<00:33, 22048.72it/s]\u001b[A\n",
      " 39%|███▉      | 466115/1193514 [00:21<00:34, 21170.23it/s]\u001b[A\n",
      " 39%|███▉      | 468428/1193514 [00:21<00:33, 21720.87it/s]\u001b[A\n",
      " 39%|███▉      | 470613/1193514 [00:21<00:34, 20863.32it/s]\u001b[A\n",
      " 40%|███▉      | 472716/1193514 [00:21<00:36, 19574.27it/s]\u001b[A\n",
      " 40%|███▉      | 475051/1193514 [00:21<00:34, 20570.56it/s]\u001b[A\n",
      " 40%|████      | 477549/1193514 [00:21<00:32, 21719.96it/s]\u001b[A\n",
      " 40%|████      | 479762/1193514 [00:21<00:32, 21790.32it/s]\u001b[A\n",
      " 40%|████      | 481970/1193514 [00:21<00:32, 21613.70it/s]\u001b[A\n",
      " 41%|████      | 484247/1193514 [00:22<00:32, 21946.77it/s]\u001b[A\n",
      " 41%|████      | 486605/1193514 [00:22<00:31, 22410.51it/s]\u001b[A\n",
      " 41%|████      | 488876/1193514 [00:22<00:31, 22497.90it/s]\u001b[A\n",
      " 41%|████      | 491339/1193514 [00:22<00:30, 23097.25it/s]\u001b[A\n",
      " 41%|████▏     | 493660/1193514 [00:22<00:33, 21126.27it/s]\u001b[A\n",
      " 42%|████▏     | 495956/1193514 [00:22<00:32, 21644.15it/s]\u001b[A\n",
      " 42%|████▏     | 498152/1193514 [00:22<00:32, 21660.06it/s]\u001b[A\n",
      " 42%|████▏     | 500584/1193514 [00:22<00:30, 22390.87it/s]\u001b[A\n",
      " 42%|████▏     | 503067/1193514 [00:22<00:29, 23070.76it/s]\u001b[A\n",
      " 42%|████▏     | 505455/1193514 [00:22<00:29, 23306.06it/s]\u001b[A\n",
      " 43%|████▎     | 507801/1193514 [00:23<00:29, 22995.89it/s]\u001b[A\n",
      " 43%|████▎     | 510257/1193514 [00:23<00:29, 23440.33it/s]\u001b[A\n",
      " 43%|████▎     | 512627/1193514 [00:23<00:28, 23516.15it/s]\u001b[A\n",
      " 43%|████▎     | 515190/1193514 [00:23<00:28, 24112.08it/s]\u001b[A\n",
      " 43%|████▎     | 517610/1193514 [00:23<00:28, 23997.73it/s]\u001b[A\n",
      " 44%|████▎     | 520016/1193514 [00:23<00:29, 23169.59it/s]\u001b[A\n",
      " 44%|████▍     | 522344/1193514 [00:23<00:30, 22324.07it/s]\u001b[A\n",
      " 44%|████▍     | 524600/1193514 [00:23<00:29, 22392.19it/s]\u001b[A\n",
      " 44%|████▍     | 527044/1193514 [00:23<00:29, 22968.44it/s]\u001b[A\n",
      " 44%|████▍     | 529517/1193514 [00:24<00:28, 23469.70it/s]\u001b[A\n",
      " 45%|████▍     | 531875/1193514 [00:24<00:28, 23358.25it/s]\u001b[A\n",
      " 45%|████▍     | 534219/1193514 [00:24<00:29, 22287.21it/s]\u001b[A\n",
      " 45%|████▍     | 536464/1193514 [00:24<00:30, 21528.36it/s]\u001b[A\n",
      " 45%|████▌     | 538915/1193514 [00:24<00:29, 22342.81it/s]\u001b[A\n",
      " 45%|████▌     | 541342/1193514 [00:24<00:28, 22886.19it/s]\u001b[A\n",
      " 46%|████▌     | 543648/1193514 [00:24<00:28, 22615.88it/s]\u001b[A\n",
      " 46%|████▌     | 545923/1193514 [00:24<00:28, 22413.16it/s]\u001b[A\n",
      " 46%|████▌     | 548307/1193514 [00:24<00:28, 22820.38it/s]\u001b[A\n",
      " 46%|████▌     | 550598/1193514 [00:24<00:28, 22678.69it/s]\u001b[A\n",
      " 46%|████▋     | 553017/1193514 [00:25<00:27, 23111.26it/s]\u001b[A\n",
      " 47%|████▋     | 555508/1193514 [00:25<00:27, 23621.28it/s]\u001b[A\n",
      " 47%|████▋     | 557878/1193514 [00:25<00:27, 23170.51it/s]\u001b[A\n",
      " 47%|████▋     | 560222/1193514 [00:25<00:27, 23249.15it/s]\u001b[A\n",
      " 47%|████▋     | 562581/1193514 [00:25<00:27, 23349.18it/s]\u001b[A\n",
      " 47%|████▋     | 564920/1193514 [00:25<00:27, 22457.00it/s]\u001b[A\n",
      " 48%|████▊     | 567176/1193514 [00:25<00:27, 22376.09it/s]\u001b[A\n",
      " 48%|████▊     | 569421/1193514 [00:25<00:29, 21377.00it/s]\u001b[A\n",
      " 48%|████▊     | 571881/1193514 [00:25<00:27, 22251.11it/s]\u001b[A\n",
      " 48%|████▊     | 574430/1193514 [00:25<00:26, 23132.18it/s]\u001b[A\n",
      " 48%|████▊     | 576784/1193514 [00:26<00:26, 23251.45it/s]\u001b[A\n",
      " 49%|████▊     | 579337/1193514 [00:26<00:25, 23889.53it/s]\u001b[A\n",
      " 49%|████▊     | 581742/1193514 [00:26<00:25, 23848.03it/s]\u001b[A\n",
      " 49%|████▉     | 584138/1193514 [00:26<00:25, 23671.95it/s]\u001b[A\n",
      " 49%|████▉     | 586514/1193514 [00:26<00:26, 22808.96it/s]\u001b[A\n",
      " 49%|████▉     | 588834/1193514 [00:26<00:26, 22922.51it/s]\u001b[A\n",
      " 50%|████▉     | 591135/1193514 [00:26<00:26, 22356.12it/s]\u001b[A\n",
      " 50%|████▉     | 593601/1193514 [00:26<00:26, 23000.20it/s]\u001b[A\n",
      " 50%|████▉     | 596093/1193514 [00:26<00:25, 23543.08it/s]\u001b[A\n",
      " 50%|█████     | 598602/1193514 [00:27<00:24, 23985.79it/s]\u001b[A\n",
      " 50%|█████     | 601097/1193514 [00:27<00:24, 24266.86it/s]\u001b[A\n",
      " 51%|█████     | 603532/1193514 [00:27<00:24, 23673.30it/s]\u001b[A\n",
      " 51%|█████     | 605908/1193514 [00:27<00:26, 22076.65it/s]\u001b[A\n",
      " 51%|█████     | 608274/1193514 [00:27<00:25, 22528.08it/s]\u001b[A\n",
      " 51%|█████     | 610706/1193514 [00:27<00:25, 23037.20it/s]\u001b[A\n",
      " 51%|█████▏    | 613028/1193514 [00:27<00:25, 23054.61it/s]\u001b[A\n",
      " 52%|█████▏    | 615486/1193514 [00:27<00:24, 23491.63it/s]\u001b[A\n",
      " 52%|█████▏    | 617900/1193514 [00:27<00:24, 23680.37it/s]\u001b[A\n",
      " 52%|█████▏    | 620486/1193514 [00:27<00:23, 24294.13it/s]\u001b[A\n",
      " 52%|█████▏    | 623009/1193514 [00:28<00:23, 24565.52it/s]\u001b[A\n",
      " 52%|█████▏    | 625505/1193514 [00:28<00:23, 24681.25it/s]\u001b[A\n",
      " 53%|█████▎    | 627979/1193514 [00:28<00:24, 23398.97it/s]\u001b[A\n",
      " 53%|█████▎    | 630337/1193514 [00:28<00:24, 23359.58it/s]\u001b[A\n",
      " 53%|█████▎    | 632686/1193514 [00:28<00:26, 21434.16it/s]\u001b[A\n",
      " 53%|█████▎    | 634921/1193514 [00:28<00:25, 21700.16it/s]\u001b[A\n",
      " 53%|█████▎    | 637135/1193514 [00:28<00:25, 21828.93it/s]\u001b[A\n",
      " 54%|█████▎    | 639339/1193514 [00:28<00:25, 21813.85it/s]\u001b[A\n",
      " 54%|█████▍    | 641728/1193514 [00:28<00:24, 22395.54it/s]\u001b[A\n",
      " 54%|█████▍    | 644199/1193514 [00:29<00:23, 23040.96it/s]\u001b[A\n",
      " 54%|█████▍    | 646777/1193514 [00:29<00:22, 23799.44it/s]\u001b[A\n",
      " 54%|█████▍    | 649174/1193514 [00:29<00:23, 23037.73it/s]\u001b[A\n",
      " 55%|█████▍    | 651495/1193514 [00:29<00:24, 21824.25it/s]\u001b[A\n",
      " 55%|█████▍    | 653703/1193514 [00:29<00:24, 21779.48it/s]\u001b[A\n",
      " 55%|█████▍    | 656118/1193514 [00:29<00:23, 22439.42it/s]\u001b[A\n",
      " 55%|█████▌    | 658380/1193514 [00:29<00:24, 22049.26it/s]\u001b[A\n",
      " 55%|█████▌    | 660599/1193514 [00:29<00:24, 21952.82it/s]\u001b[A\n",
      " 56%|█████▌    | 662840/1193514 [00:29<00:24, 22087.90it/s]\u001b[A\n",
      " 56%|█████▌    | 665117/1193514 [00:29<00:23, 22287.66it/s]\u001b[A\n",
      " 56%|█████▌    | 667404/1193514 [00:30<00:23, 22459.15it/s]\u001b[A\n",
      " 56%|█████▌    | 669775/1193514 [00:30<00:22, 22818.13it/s]\u001b[A\n",
      " 56%|█████▋    | 672112/1193514 [00:30<00:22, 22979.91it/s]\u001b[A\n",
      " 57%|█████▋    | 674414/1193514 [00:30<00:24, 21581.20it/s]\u001b[A\n",
      " 57%|█████▋    | 676788/1193514 [00:30<00:23, 22185.59it/s]\u001b[A\n",
      " 57%|█████▋    | 679025/1193514 [00:30<00:24, 20677.37it/s]\u001b[A\n",
      " 57%|█████▋    | 681450/1193514 [00:30<00:23, 21632.83it/s]\u001b[A\n",
      " 57%|█████▋    | 683649/1193514 [00:30<00:25, 20264.58it/s]\u001b[A\n",
      " 57%|█████▋    | 685834/1193514 [00:30<00:24, 20714.97it/s]\u001b[A\n",
      " 58%|█████▊    | 688191/1193514 [00:31<00:23, 21493.65it/s]\u001b[A\n",
      " 58%|█████▊    | 690385/1193514 [00:31<00:23, 21623.62it/s]\u001b[A\n",
      " 58%|█████▊    | 692569/1193514 [00:31<00:24, 20862.92it/s]\u001b[A\n",
      " 58%|█████▊    | 694676/1193514 [00:31<00:25, 19466.11it/s]\u001b[A\n",
      " 58%|█████▊    | 696929/1193514 [00:31<00:24, 20292.68it/s]\u001b[A\n",
      " 59%|█████▊    | 699224/1193514 [00:31<00:23, 21021.28it/s]\u001b[A\n",
      " 59%|█████▉    | 701531/1193514 [00:31<00:22, 21594.95it/s]\u001b[A\n",
      " 59%|█████▉    | 703715/1193514 [00:31<00:22, 21450.92it/s]\u001b[A\n",
      " 59%|█████▉    | 705878/1193514 [00:31<00:22, 21464.30it/s]\u001b[A\n",
      " 59%|█████▉    | 708037/1193514 [00:31<00:23, 20576.53it/s]\u001b[A\n",
      " 60%|█████▉    | 710432/1193514 [00:32<00:22, 21482.80it/s]\u001b[A\n",
      " 60%|█████▉    | 712804/1193514 [00:32<00:21, 22107.83it/s]\u001b[A\n",
      " 60%|█████▉    | 715097/1193514 [00:32<00:21, 22348.06it/s]\u001b[A\n",
      " 60%|██████    | 717347/1193514 [00:32<00:21, 22027.80it/s]\u001b[A\n",
      " 60%|██████    | 719577/1193514 [00:32<00:21, 22107.40it/s]\u001b[A\n",
      " 60%|██████    | 721954/1193514 [00:32<00:20, 22579.10it/s]\u001b[A\n",
      " 61%|██████    | 724321/1193514 [00:32<00:20, 22892.74it/s]\u001b[A\n",
      " 61%|██████    | 726751/1193514 [00:32<00:20, 23297.28it/s]\u001b[A\n",
      " 61%|██████    | 729088/1193514 [00:32<00:19, 23228.56it/s]\u001b[A\n",
      " 61%|██████▏   | 731416/1193514 [00:32<00:20, 22673.28it/s]\u001b[A\n",
      " 61%|██████▏   | 733690/1193514 [00:33<00:20, 22436.64it/s]\u001b[A\n",
      " 62%|██████▏   | 735939/1193514 [00:33<00:20, 22406.32it/s]\u001b[A\n",
      " 62%|██████▏   | 738184/1193514 [00:33<00:21, 21670.17it/s]\u001b[A\n",
      " 62%|██████▏   | 740359/1193514 [00:33<00:22, 19879.88it/s]\u001b[A\n",
      " 62%|██████▏   | 742720/1193514 [00:33<00:21, 20867.46it/s]\u001b[A\n",
      " 62%|██████▏   | 744844/1193514 [00:33<00:21, 20403.36it/s]\u001b[A\n",
      " 63%|██████▎   | 747351/1193514 [00:33<00:20, 21609.07it/s]\u001b[A\n",
      " 63%|██████▎   | 749551/1193514 [00:33<00:21, 21129.86it/s]\u001b[A\n",
      " 63%|██████▎   | 751694/1193514 [00:33<00:21, 20643.56it/s]\u001b[A\n",
      " 63%|██████▎   | 754119/1193514 [00:34<00:20, 21605.62it/s]\u001b[A\n",
      " 63%|██████▎   | 756575/1193514 [00:34<00:19, 22412.64it/s]\u001b[A\n",
      " 64%|██████▎   | 758844/1193514 [00:34<00:20, 20825.10it/s]\u001b[A\n",
      " 64%|██████▍   | 761134/1193514 [00:34<00:20, 21404.93it/s]\u001b[A\n",
      " 64%|██████▍   | 763308/1193514 [00:34<00:20, 20907.69it/s]\u001b[A\n",
      " 64%|██████▍   | 765668/1193514 [00:34<00:19, 21647.23it/s]\u001b[A\n",
      " 64%|██████▍   | 768009/1193514 [00:34<00:19, 22144.88it/s]\u001b[A\n",
      " 65%|██████▍   | 770263/1193514 [00:34<00:19, 22260.89it/s]\u001b[A\n",
      " 65%|██████▍   | 772728/1193514 [00:34<00:18, 22926.98it/s]\u001b[A\n",
      " 65%|██████▍   | 775100/1193514 [00:35<00:18, 23158.80it/s]\u001b[A\n",
      " 65%|██████▌   | 777427/1193514 [00:35<00:18, 22640.60it/s]\u001b[A\n",
      " 65%|██████▌   | 779702/1193514 [00:35<00:18, 22055.41it/s]\u001b[A\n",
      " 66%|██████▌   | 782168/1193514 [00:35<00:18, 22774.93it/s]\u001b[A\n",
      " 66%|██████▌   | 784459/1193514 [00:35<00:18, 21752.20it/s]\u001b[A\n",
      " 66%|██████▌   | 786754/1193514 [00:35<00:18, 22096.72it/s]\u001b[A\n",
      " 66%|██████▌   | 788979/1193514 [00:35<00:18, 21411.41it/s]\u001b[A\n",
      " 66%|██████▋   | 791253/1193514 [00:35<00:18, 21793.34it/s]\u001b[A\n",
      " 67%|██████▋   | 793698/1193514 [00:35<00:17, 22526.88it/s]\u001b[A\n",
      " 67%|██████▋   | 795966/1193514 [00:35<00:18, 21901.13it/s]\u001b[A\n",
      " 67%|██████▋   | 798178/1193514 [00:36<00:17, 21965.59it/s]\u001b[A\n",
      " 67%|██████▋   | 800385/1193514 [00:36<00:17, 21977.80it/s]\u001b[A\n",
      " 67%|██████▋   | 802700/1193514 [00:36<00:17, 22315.50it/s]\u001b[A\n",
      " 67%|██████▋   | 805103/1193514 [00:36<00:17, 22803.11it/s]\u001b[A\n",
      " 68%|██████▊   | 807391/1193514 [00:36<00:17, 22416.37it/s]\u001b[A\n",
      " 68%|██████▊   | 809639/1193514 [00:36<00:17, 21565.51it/s]\u001b[A\n",
      " 68%|██████▊   | 811981/1193514 [00:36<00:17, 22090.05it/s]\u001b[A\n",
      " 68%|██████▊   | 814401/1193514 [00:36<00:16, 22681.59it/s]\u001b[A\n",
      " 68%|██████▊   | 816927/1193514 [00:36<00:16, 23396.25it/s]\u001b[A\n",
      " 69%|██████▊   | 819281/1193514 [00:36<00:16, 22849.69it/s]\u001b[A\n",
      " 69%|██████▉   | 821713/1193514 [00:37<00:15, 23270.46it/s]\u001b[A\n",
      " 69%|██████▉   | 824051/1193514 [00:37<00:16, 21815.40it/s]\u001b[A\n",
      " 69%|██████▉   | 826492/1193514 [00:37<00:16, 22533.76it/s]\u001b[A\n",
      " 69%|██████▉   | 828770/1193514 [00:37<00:16, 21867.69it/s]\u001b[A\n",
      " 70%|██████▉   | 830978/1193514 [00:37<00:16, 21788.84it/s]\u001b[A\n",
      " 70%|██████▉   | 833299/1193514 [00:37<00:16, 22194.14it/s]\u001b[A\n",
      " 70%|███████   | 835667/1193514 [00:37<00:15, 22619.17it/s]\u001b[A\n",
      " 70%|███████   | 837940/1193514 [00:37<00:15, 22467.02it/s]\u001b[A\n",
      " 70%|███████   | 840195/1193514 [00:37<00:16, 21023.05it/s]\u001b[A\n",
      " 71%|███████   | 842641/1193514 [00:38<00:15, 21947.39it/s]\u001b[A\n",
      " 71%|███████   | 845064/1193514 [00:38<00:15, 22584.46it/s]\u001b[A\n",
      " 71%|███████   | 847493/1193514 [00:38<00:15, 23067.30it/s]\u001b[A\n",
      " 71%|███████   | 849965/1193514 [00:38<00:14, 23539.01it/s]\u001b[A\n",
      " 71%|███████▏  | 852335/1193514 [00:38<00:15, 22468.91it/s]\u001b[A\n",
      " 72%|███████▏  | 854729/1193514 [00:38<00:14, 22890.28it/s]\u001b[A\n",
      " 72%|███████▏  | 857035/1193514 [00:38<00:14, 22705.19it/s]\u001b[A\n",
      " 72%|███████▏  | 859389/1193514 [00:38<00:14, 22948.10it/s]\u001b[A\n",
      " 72%|███████▏  | 861693/1193514 [00:38<00:14, 22592.65it/s]\u001b[A\n",
      " 72%|███████▏  | 863960/1193514 [00:38<00:14, 22252.13it/s]\u001b[A\n",
      " 73%|███████▎  | 866349/1193514 [00:39<00:14, 22717.91it/s]\u001b[A\n",
      " 73%|███████▎  | 868644/1193514 [00:39<00:14, 22785.29it/s]\u001b[A\n",
      " 73%|███████▎  | 871057/1193514 [00:39<00:13, 23172.00it/s]\u001b[A\n",
      " 73%|███████▎  | 873469/1193514 [00:39<00:13, 23447.72it/s]\u001b[A\n",
      " 73%|███████▎  | 875819/1193514 [00:39<00:14, 22461.95it/s]\u001b[A\n",
      " 74%|███████▎  | 878153/1193514 [00:39<00:13, 22716.36it/s]\u001b[A\n",
      " 74%|███████▍  | 880434/1193514 [00:39<00:14, 22027.65it/s]\u001b[A\n",
      " 74%|███████▍  | 882675/1193514 [00:39<00:14, 22140.30it/s]\u001b[A\n",
      " 74%|███████▍  | 885204/1193514 [00:39<00:13, 22999.33it/s]\u001b[A\n",
      " 74%|███████▍  | 887518/1193514 [00:40<00:14, 21670.87it/s]\u001b[A\n",
      " 75%|███████▍  | 889898/1193514 [00:40<00:13, 22266.89it/s]\u001b[A\n",
      " 75%|███████▍  | 892327/1193514 [00:40<00:13, 22834.42it/s]\u001b[A\n",
      " 75%|███████▍  | 894702/1193514 [00:40<00:12, 23100.57it/s]\u001b[A\n",
      " 75%|███████▌  | 897075/1193514 [00:40<00:12, 23281.34it/s]\u001b[A\n",
      " 75%|███████▌  | 899414/1193514 [00:40<00:13, 22594.43it/s]\u001b[A\n",
      " 76%|███████▌  | 901685/1193514 [00:40<00:12, 22453.68it/s]\u001b[A\n",
      " 76%|███████▌  | 904078/1193514 [00:40<00:12, 22874.68it/s]\u001b[A\n",
      " 76%|███████▌  | 906434/1193514 [00:40<00:12, 23074.57it/s]\u001b[A\n",
      " 76%|███████▌  | 908872/1193514 [00:40<00:12, 23449.08it/s]\u001b[A\n",
      " 76%|███████▋  | 911223/1193514 [00:41<00:12, 22219.22it/s]\u001b[A\n",
      " 77%|███████▋  | 913588/1193514 [00:41<00:12, 22626.57it/s]\u001b[A\n",
      " 77%|███████▋  | 915865/1193514 [00:41<00:12, 22565.26it/s]\u001b[A\n",
      " 77%|███████▋  | 918132/1193514 [00:41<00:12, 22506.33it/s]\u001b[A\n",
      " 77%|███████▋  | 920392/1193514 [00:41<00:12, 22533.73it/s]\u001b[A\n",
      " 77%|███████▋  | 922710/1193514 [00:41<00:11, 22717.46it/s]\u001b[A\n",
      " 78%|███████▊  | 924986/1193514 [00:41<00:11, 22582.21it/s]\u001b[A\n",
      " 78%|███████▊  | 927248/1193514 [00:41<00:12, 22024.83it/s]\u001b[A\n",
      " 78%|███████▊  | 929456/1193514 [00:41<00:13, 19998.14it/s]\u001b[A\n",
      " 78%|███████▊  | 931496/1193514 [00:42<00:13, 19985.32it/s]\u001b[A\n",
      " 78%|███████▊  | 933893/1193514 [00:42<00:12, 21031.52it/s]\u001b[A\n",
      " 78%|███████▊  | 936030/1193514 [00:42<00:12, 20909.09it/s]\u001b[A\n",
      " 79%|███████▊  | 938372/1193514 [00:42<00:11, 21603.62it/s]\u001b[A\n",
      " 79%|███████▉  | 940743/1193514 [00:42<00:11, 22193.19it/s]\u001b[A\n",
      " 79%|███████▉  | 942983/1193514 [00:42<00:11, 21614.55it/s]\u001b[A\n",
      " 79%|███████▉  | 945175/1193514 [00:42<00:11, 21703.14it/s]\u001b[A\n",
      " 79%|███████▉  | 947483/1193514 [00:42<00:11, 22096.75it/s]\u001b[A\n",
      " 80%|███████▉  | 949710/1193514 [00:42<00:11, 22145.34it/s]\u001b[A\n",
      " 80%|███████▉  | 952032/1193514 [00:42<00:10, 22455.76it/s]\u001b[A\n",
      " 80%|███████▉  | 954313/1193514 [00:43<00:10, 22559.15it/s]\u001b[A\n",
      " 80%|████████  | 956664/1193514 [00:43<00:10, 22835.73it/s]\u001b[A\n",
      " 80%|████████  | 959078/1193514 [00:43<00:10, 23210.49it/s]\u001b[A\n",
      " 81%|████████  | 961404/1193514 [00:43<00:10, 23077.31it/s]\u001b[A\n",
      " 81%|████████  | 963715/1193514 [00:43<00:10, 22526.64it/s]\u001b[A\n",
      " 81%|████████  | 965973/1193514 [00:43<00:10, 21294.57it/s]\u001b[A\n",
      " 81%|████████  | 968202/1193514 [00:43<00:10, 21583.31it/s]\u001b[A\n",
      " 81%|████████▏ | 970374/1193514 [00:43<00:10, 20406.17it/s]\u001b[A\n",
      " 81%|████████▏ | 972644/1193514 [00:43<00:10, 21043.58it/s]\u001b[A\n",
      " 82%|████████▏ | 974770/1193514 [00:43<00:10, 20743.03it/s]\u001b[A\n",
      " 82%|████████▏ | 976975/1193514 [00:44<00:10, 21116.58it/s]\u001b[A\n",
      " 82%|████████▏ | 979100/1193514 [00:44<00:10, 20387.72it/s]\u001b[A\n",
      " 82%|████████▏ | 981391/1193514 [00:44<00:10, 21082.55it/s]\u001b[A\n",
      " 82%|████████▏ | 983717/1193514 [00:44<00:09, 21691.47it/s]\u001b[A\n",
      " 83%|████████▎ | 985902/1193514 [00:44<00:09, 21434.75it/s]\u001b[A\n",
      " 83%|████████▎ | 988058/1193514 [00:44<00:09, 21086.74it/s]\u001b[A\n",
      " 83%|████████▎ | 990177/1193514 [00:44<00:09, 20531.90it/s]\u001b[A\n",
      " 83%|████████▎ | 992403/1193514 [00:44<00:09, 21021.47it/s]\u001b[A\n",
      " 83%|████████▎ | 994741/1193514 [00:44<00:09, 21675.72it/s]\u001b[A\n",
      " 84%|████████▎ | 996921/1193514 [00:45<00:09, 21139.17it/s]\u001b[A\n",
      " 84%|████████▎ | 999047/1193514 [00:45<00:09, 20353.24it/s]\u001b[A\n",
      " 84%|████████▍ | 1001097/1193514 [00:45<00:09, 19703.09it/s]\u001b[A\n",
      " 84%|████████▍ | 1003416/1193514 [00:45<00:09, 20633.56it/s]\u001b[A\n",
      " 84%|████████▍ | 1005501/1193514 [00:45<00:09, 20498.57it/s]\u001b[A\n",
      " 84%|████████▍ | 1007742/1193514 [00:45<00:08, 21036.16it/s]\u001b[A\n",
      " 85%|████████▍ | 1009860/1193514 [00:45<00:09, 20127.21it/s]\u001b[A\n",
      " 85%|████████▍ | 1012194/1193514 [00:45<00:08, 20992.18it/s]\u001b[A\n",
      " 85%|████████▌ | 1014616/1193514 [00:45<00:08, 21864.71it/s]\u001b[A\n",
      " 85%|████████▌ | 1016828/1193514 [00:45<00:08, 21868.40it/s]\u001b[A\n",
      " 85%|████████▌ | 1019033/1193514 [00:46<00:08, 21473.02it/s]\u001b[A\n",
      " 86%|████████▌ | 1021359/1193514 [00:46<00:07, 21978.60it/s]\u001b[A\n",
      " 86%|████████▌ | 1023801/1193514 [00:46<00:07, 22656.02it/s]\u001b[A\n",
      " 86%|████████▌ | 1026218/1193514 [00:46<00:07, 23088.17it/s]\u001b[A\n",
      " 86%|████████▌ | 1028539/1193514 [00:46<00:07, 22120.35it/s]\u001b[A\n",
      " 86%|████████▋ | 1030768/1193514 [00:46<00:07, 21388.51it/s]\u001b[A\n",
      " 87%|████████▋ | 1032924/1193514 [00:46<00:07, 21320.59it/s]\u001b[A\n",
      " 87%|████████▋ | 1035367/1193514 [00:46<00:07, 22161.23it/s]\u001b[A\n",
      " 87%|████████▋ | 1037601/1193514 [00:46<00:07, 21161.05it/s]\u001b[A\n",
      " 87%|████████▋ | 1039798/1193514 [00:47<00:07, 21393.38it/s]\u001b[A\n",
      " 87%|████████▋ | 1042048/1193514 [00:47<00:06, 21713.38it/s]\u001b[A\n",
      " 88%|████████▊ | 1044559/1193514 [00:47<00:06, 22630.92it/s]\u001b[A\n",
      " 88%|████████▊ | 1046841/1193514 [00:47<00:06, 21737.31it/s]\u001b[A\n",
      " 88%|████████▊ | 1049083/1193514 [00:47<00:06, 21936.12it/s]\u001b[A\n",
      " 88%|████████▊ | 1051321/1193514 [00:47<00:06, 22065.25it/s]\u001b[A\n",
      " 88%|████████▊ | 1053539/1193514 [00:47<00:06, 21498.37it/s]\u001b[A\n",
      " 88%|████████▊ | 1055700/1193514 [00:47<00:06, 20872.54it/s]\u001b[A\n",
      " 89%|████████▊ | 1057799/1193514 [00:47<00:06, 20844.34it/s]\u001b[A\n",
      " 89%|████████▉ | 1060269/1193514 [00:47<00:06, 21866.70it/s]\u001b[A\n",
      " 89%|████████▉ | 1062773/1193514 [00:48<00:05, 22730.03it/s]\u001b[A\n",
      " 89%|████████▉ | 1065142/1193514 [00:48<00:05, 23009.34it/s]\u001b[A\n",
      " 89%|████████▉ | 1067460/1193514 [00:48<00:05, 22461.49it/s]\u001b[A\n",
      " 90%|████████▉ | 1069918/1193514 [00:48<00:05, 23055.72it/s]\u001b[A\n",
      " 90%|████████▉ | 1072238/1193514 [00:48<00:05, 21851.57it/s]\u001b[A\n",
      " 90%|█████████ | 1074446/1193514 [00:48<00:05, 20051.59it/s]\u001b[A\n",
      " 90%|█████████ | 1076729/1193514 [00:48<00:05, 20810.89it/s]\u001b[A\n",
      " 90%|█████████ | 1079256/1193514 [00:48<00:05, 21973.77it/s]\u001b[A\n",
      " 91%|█████████ | 1081498/1193514 [00:48<00:05, 21792.45it/s]\u001b[A\n",
      " 91%|█████████ | 1083740/1193514 [00:49<00:04, 21970.79it/s]\u001b[A\n",
      " 91%|█████████ | 1086144/1193514 [00:49<00:04, 22551.28it/s]\u001b[A\n",
      " 91%|█████████ | 1088564/1193514 [00:49<00:04, 23019.70it/s]\u001b[A\n",
      " 91%|█████████▏| 1090883/1193514 [00:49<00:04, 21930.21it/s]\u001b[A\n",
      " 92%|█████████▏| 1093266/1193514 [00:49<00:04, 22467.23it/s]\u001b[A\n",
      " 92%|█████████▏| 1095532/1193514 [00:49<00:04, 21898.11it/s]\u001b[A\n",
      " 92%|█████████▏| 1097740/1193514 [00:49<00:04, 21951.21it/s]\u001b[A\n",
      " 92%|█████████▏| 1100177/1193514 [00:49<00:04, 22622.86it/s]\u001b[A\n",
      " 92%|█████████▏| 1102584/1193514 [00:49<00:03, 23038.23it/s]\u001b[A\n",
      " 93%|█████████▎| 1104923/1193514 [00:49<00:03, 23142.01it/s]\u001b[A\n",
      " 93%|█████████▎| 1107416/1193514 [00:50<00:03, 23647.88it/s]\u001b[A\n",
      " 93%|█████████▎| 1109790/1193514 [00:50<00:03, 23532.08it/s]\u001b[A\n",
      " 93%|█████████▎| 1112150/1193514 [00:50<00:03, 22960.59it/s]\u001b[A\n",
      " 93%|█████████▎| 1114606/1193514 [00:50<00:03, 23416.93it/s]\u001b[A\n",
      " 94%|█████████▎| 1116955/1193514 [00:50<00:03, 22845.05it/s]\u001b[A\n",
      " 94%|█████████▍| 1119262/1193514 [00:50<00:03, 22907.91it/s]\u001b[A\n",
      " 94%|█████████▍| 1121559/1193514 [00:50<00:03, 22443.12it/s]\u001b[A\n",
      " 94%|█████████▍| 1124014/1193514 [00:50<00:03, 23034.69it/s]\u001b[A\n",
      " 94%|█████████▍| 1126488/1193514 [00:50<00:02, 23519.58it/s]\u001b[A\n",
      " 95%|█████████▍| 1129008/1193514 [00:50<00:02, 23999.37it/s]\u001b[A\n",
      " 95%|█████████▍| 1131417/1193514 [00:51<00:02, 23853.03it/s]\u001b[A\n",
      " 95%|█████████▍| 1133809/1193514 [00:51<00:02, 23829.82it/s]\u001b[A\n",
      " 95%|█████████▌| 1136299/1193514 [00:51<00:02, 24141.08it/s]\u001b[A\n",
      " 95%|█████████▌| 1138769/1193514 [00:51<00:02, 24305.49it/s]\u001b[A\n",
      " 96%|█████████▌| 1141203/1193514 [00:51<00:02, 23885.35it/s]\u001b[A\n",
      " 96%|█████████▌| 1143596/1193514 [00:51<00:02, 22785.11it/s]\u001b[A\n",
      " 96%|█████████▌| 1146121/1193514 [00:51<00:02, 23472.46it/s]\u001b[A\n",
      " 96%|█████████▌| 1148552/1193514 [00:51<00:01, 23716.22it/s]\u001b[A\n",
      " 96%|█████████▋| 1150935/1193514 [00:51<00:01, 22828.89it/s]\u001b[A\n",
      " 97%|█████████▋| 1153309/1193514 [00:52<00:01, 23093.13it/s]\u001b[A\n",
      " 97%|█████████▋| 1155783/1193514 [00:52<00:01, 23563.00it/s]\u001b[A\n",
      " 97%|█████████▋| 1158150/1193514 [00:52<00:01, 22622.38it/s]\u001b[A\n",
      " 97%|█████████▋| 1160428/1193514 [00:52<00:01, 21849.17it/s]\u001b[A\n",
      " 97%|█████████▋| 1162630/1193514 [00:52<00:01, 21252.34it/s]\u001b[A\n",
      " 98%|█████████▊| 1165013/1193514 [00:52<00:01, 21964.04it/s]\u001b[A\n",
      " 98%|█████████▊| 1167226/1193514 [00:52<00:01, 22011.63it/s]\u001b[A\n",
      " 98%|█████████▊| 1169439/1193514 [00:52<00:01, 21416.07it/s]\u001b[A\n",
      " 98%|█████████▊| 1171593/1193514 [00:52<00:01, 20325.01it/s]\u001b[A\n",
      " 98%|█████████▊| 1174029/1193514 [00:52<00:00, 21387.31it/s]\u001b[A\n",
      " 99%|█████████▊| 1176365/1193514 [00:53<00:00, 21941.41it/s]\u001b[A\n",
      " 99%|█████████▉| 1178795/1193514 [00:53<00:00, 22597.66it/s]\u001b[A\n",
      " 99%|█████████▉| 1181077/1193514 [00:53<00:00, 22144.24it/s]\u001b[A\n",
      " 99%|█████████▉| 1183452/1193514 [00:53<00:00, 22601.20it/s]\u001b[A\n",
      " 99%|█████████▉| 1185909/1193514 [00:53<00:00, 23156.43it/s]\u001b[A\n",
      "100%|█████████▉| 1188348/1193514 [00:53<00:00, 23512.17it/s]\u001b[A\n",
      "100%|█████████▉| 1190710/1193514 [00:53<00:00, 22637.43it/s]\u001b[A\n",
      "100%|█████████▉| 1192989/1193514 [00:53<00:00, 22027.51it/s]\u001b[A"
     ]
    }
   ],
   "source": [
    "# Or uploading the file directly from Drive\n",
    "glove_25 = vocab.Vectors(r\"/content/drive/My Drive/glove.twitter.27B.25d.txt\")\n",
    "glove_100 = vocab.Vectors(r\"/content/drive/My Drive/glove.twitter.27B.100d.txt\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "y0IPIFAoptUh"
   },
   "source": [
    "## Creating __GloVe vocabulary__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "aW4J5rYpkScY"
   },
   "source": [
    "Cleaning dataframe from unused columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "q6jem6hiauTD"
   },
   "outputs": [],
   "source": [
    "tweetDf.drop(columns=['tweet','text'],inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Lcr5cq7_fzqk"
   },
   "source": [
    "Creating Fields for my text column and the targets. After I am building a vocabulary from the GloVe embeddings and the tweets. This vocabulary has:\n",
    "- All words from the tweets (uniquely)\n",
    "- All words from the tweets (uniquely) that belong also to GloVe embeddings.\n",
    "\n",
    "These words are corellated to a vector (25 dimension) from Glove or with <pad>,<unk>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "QIvyWap6Q9fC"
   },
   "source": [
    "### Data dimensions\n",
    "```\n",
    "vector_length # after vectorization in the size of the vector that will be passed to the nets\n",
    "glove_length  # the length of the GloVe vector given from the Stanford files\n",
    "glove_package # the Stanford package\n",
    "\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "4KYbNJ2Sze59"
   },
   "outputs": [],
   "source": [
    "vector_length = 100\n",
    "glove_length  = 25\n",
    "glove_package = glove_25"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "mxX6cz_YOdxP"
   },
   "outputs": [],
   "source": [
    "text_field = Field(tokenize=nltk.word_tokenize,lower=True,fix_length = vector_length,sequential=True) # tokenize text using word_tokenize and convert to numerical form using default parameters\n",
    "label_field = Field(sequential=False, use_vocab=False)\n",
    "\n",
    "fields = [('target',label_field),('processed_tweets',text_field)]\n",
    "\n",
    "preprocessed_text = tweetDf['processed_tweets'].apply(lambda x: text_field.preprocess(x))\n",
    "text_field.build_vocab(preprocessed_text,vectors=glove_package)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "uHMibRw0mex6"
   },
   "source": [
    "### Creating datasets\n",
    "\n",
    "As examples I give all words from column 'processed_tweets' of the dataframe. Sentences are tokenized from Field function with word_tokenizer from nltk. Also the new sets have sizes:\n",
    "- Train 80%\n",
    "- Validation 10%\n",
    "- Test 10%\n",
    "\n",
    "\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "ttfL0t_8qlon",
    "outputId": "c2c26a8e-0ec1-40b1-b1d1-26f095ac49e3"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n",
      "100%|█████████▉| 1192989/1193514 [01:09<00:00, 22027.51it/s]\u001b[A"
     ]
    }
   ],
   "source": [
    "from torchtext.data import Dataset, Example\n",
    "train_dataset,validation_dataset,test_dataset = Dataset(\n",
    "    examples=[Example.fromlist(list(r),fields) for i,r in tweetDf.iterrows()],\n",
    "    fields=fields).split(split_ratio=[0.8,0.1,0.1])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "tJEWUlnEmr56"
   },
   "source": [
    "__Batch size__ : I tried many batche-sizes (32,128,256) but I concluded to this size."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "Pq5qqIMYenAG"
   },
   "outputs": [],
   "source": [
    "batch_size=128"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "OUD2-7t7myhA"
   },
   "source": [
    "### Creating batches\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "-TScledPAJfS"
   },
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "T67BBcf5jLlR"
   },
   "outputs": [],
   "source": [
    "train_loader, valid_loader, test_loader = BucketIterator.splits(\n",
    "    datasets=(train_dataset,validation_dataset ,test_dataset), \n",
    "    batch_sizes=(batch_size,batch_size,batch_size),\n",
    "    sort=False,\n",
    "    device = 'cuda' if train_on_gpu else 'cpu',\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "4g66u1IinG_w"
   },
   "source": [
    "# Neural Network __functions and utilities__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "SICGpkaynuET"
   },
   "source": [
    "## __Training__ and __validation__ function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "E-8g0xWpnwfM"
   },
   "outputs": [],
   "source": [
    "def train_and_validate(net,epochs,loss_function,optimizer,scheduler,args):\n",
    "\n",
    "  train_loss, train_acc = [], []\n",
    "  val_loss_list, val_acc_list = [], []\n",
    "\n",
    "  # number of epochs\n",
    "  if(train_on_gpu):\n",
    "    net.cuda()\n",
    "\n",
    "  for e in range(epochs):\n",
    "    \n",
    "    running_loss = 0.0\n",
    "    running_acc = 0.0\n",
    "    counter = 0\n",
    "    val_loss = 0.0\n",
    "    val_acc = 0.0\n",
    "    \n",
    "    # ------------- TRAIN --------------\n",
    "    net.train()    \n",
    "    for batch in train_loader:\n",
    "        counter += 1\n",
    "        \n",
    "        if(train_on_gpu):\n",
    "            batch.processed_tweets, batch.target = batch.processed_tweets.cuda(), batch.target.cuda()\n",
    "\n",
    "        optimizer.zero_grad()\n",
    "        # print(\"-- \",batch.processed_tweets.shape)\n",
    "        y_pred = net(batch.processed_tweets.T)\n",
    "\n",
    "        if(train_on_gpu):\n",
    "          y_pred = y_pred.squeeze().cuda()\n",
    "          target = batch.target.float().cuda()\n",
    "        else:\n",
    "          y_pred = y_pred.squeeze()\n",
    "          target = batch.target.float()\n",
    "\n",
    "        l = loss_function(y_pred, target)\n",
    "        running_loss += (l.item() - running_loss) / (counter)\n",
    "\n",
    "        l.backward()\n",
    "\n",
    "        # Clip gradients\n",
    "        if args.gradient_clip != None:\n",
    "          for param in net.parameters():\n",
    "              if param.grad is None:\n",
    "                  continue\n",
    "              grad_val = torch.clamp(param.grad, -args.gradient_clip, args.gradient_clip)\n",
    "\n",
    "        # apply gradients\n",
    "        optimizer.step()\n",
    "\n",
    "        acc_t = accuracy_score(batch.target.cpu().numpy(),torch.round(y_pred).cpu().detach().numpy())*100\n",
    "        running_acc += (acc_t - running_acc) / (counter)\n",
    "\n",
    "\n",
    "    train_loss.append(running_loss)\n",
    "    train_acc.append(running_acc)\n",
    "\n",
    "    # -------------- VALIDATE --------------\n",
    "\n",
    "    epoch_loss = 0.0\n",
    "    counter = 0\n",
    "    with torch.no_grad():\n",
    "        # Faster inference\n",
    "        net.eval()\n",
    "\n",
    "\n",
    "    for batch in valid_loader:\n",
    "      \n",
    "      counter += 1\n",
    "\n",
    "      if(train_on_gpu):\n",
    "        batch.processed_tweets, batch.target = batch.processed_tweets.cuda(), batch.target.cuda()\n",
    "  \n",
    "      y_pred = net.forward(batch.processed_tweets.T)\n",
    "\n",
    "      if(train_on_gpu):\n",
    "        y_pred = y_pred.squeeze().cuda()\n",
    "        target = batch.target.float().cuda()\n",
    "      else:\n",
    "        y_pred = y_pred.squeeze()\n",
    "        target = batch.target.float()\n",
    "\n",
    "      l = loss_function(y_pred, target)\n",
    "\n",
    "      val_loss += (l.item() - val_loss) / (counter)\n",
    "      # y_pred = torch.round(y_pred)\n",
    "      acc_t = accuracy_score(batch.target.cpu().numpy(),torch.round(y_pred).cpu().detach().numpy())*100\n",
    "      val_acc += (acc_t - val_acc) / (counter)\n",
    "\n",
    "    val_loss_list.append(val_loss)\n",
    "    val_acc_list.append(val_acc)\n",
    "    if scheduler != None:\n",
    "      scheduler.step(val_loss_list[-1])\n",
    "    print(f'Epoch {(e+1)+0:03} | Train_Loss: {running_loss:.5f} | Train_Acc: {running_acc:.3f} | Val_Loss: {val_loss:.3f} | Val_Acc: {val_acc:.3f}')\n",
    "  \n",
    "  print()\n",
    "  print(\"TRAIN STATS:\")\n",
    "  print(\"Train mean loss: {:.6f} \".format(np.mean(train_loss)))\n",
    "  print(\"Train mean acc: {:.6f}\".format(np.mean(train_acc)))\n",
    "  print()\n",
    "  print(\"VALIDATION STATS:\")\n",
    "  print(\"Validation mean loss: {:.6f} \".format(np.mean(val_loss)))\n",
    "  print(\"Validation mean acc: {:.6f}\".format(np.mean(val_acc)))     \n",
    "\n",
    "  return train_loss,train_acc,val_loss_list,val_acc_list"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "lXB6WpcLn4Py"
   },
   "source": [
    "## __Test__ function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "mhXE30GlKCCX"
   },
   "outputs": [],
   "source": [
    "def test(net,test_loader,loss_function):\n",
    "  running_loss = 0.0\n",
    "  score_accuracy = []\n",
    "  score_f1 = []\n",
    "  score_precision = []\n",
    "  score_recall = []\n",
    "  counter = 0\n",
    "\n",
    "  net.eval()\n",
    "  for batch in test_loader:\n",
    "      counter += 1\n",
    "\n",
    "      if(train_on_gpu):\n",
    "        batch.processed_tweets, batch.target = batch.processed_tweets.cuda(), batch.target.cuda()\n",
    "  \n",
    "      y_pred = net.forward(batch.processed_tweets.T)\n",
    "\n",
    "      if(train_on_gpu):\n",
    "        batch.processed_tweets, batch.target = batch.processed_tweets.cpu(), batch.target.cpu()\n",
    "\n",
    "      y_pred = torch.round(y_pred).cpu()\n",
    "      score_f1.append(f1_score(batch.target.numpy(),y_pred.detach().numpy()))\n",
    "      score_accuracy.append(accuracy_score(batch.target.numpy(),y_pred.detach().numpy()))\n",
    "      score_recall.append(recall_score(batch.target.numpy(),y_pred.detach().numpy()))\n",
    "      score_precision.append(precision_score(batch.target.numpy(),y_pred.detach().numpy()))\n",
    "      \n",
    "  print(\"F1:        {:.6f}\".format(np.mean(score_f1)*100))\n",
    "  print(\"Recall:    {:.6f}\".format(np.mean(score_recall)*100))\n",
    "  print(\"Precision: {:.6f}\".format(np.mean(score_precision)*100))\n",
    "  print(\"Accuracy:  {:.6f}\".format(np.mean(score_accuracy)*100))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "ddYyETUEoTkW"
   },
   "source": [
    "## Custom __Grid search__ function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "EBm3Q3uLoWCI"
   },
   "outputs": [],
   "source": [
    "from torch.optim import Adam\n",
    "\n",
    "def grid_search(optimizers,learning_rates,hidden_layers):\n",
    "\n",
    "    counter = 0\n",
    "    for optimizer_name in optimizers:\n",
    "      for lr in learning_rates:\n",
    "        for hidden_layer in hidden_layers:\n",
    "          counter += 1\n",
    "          print(\"----------------------------------------------\")\n",
    "          print(\"\\n Grid Search \",counter,\"\\n- Optimizer:  \",str(optimizer_name),\"\\n- Learning rate: \",lr,\"\\n- Hidden layers: \",hidden_layer,\"\\n\\n\")\n",
    "          epochs = 5\n",
    "          loss_function = nn.BCELoss()\n",
    "          net = TweetClassifier(vocab_size = len(text_field.vocab),\n",
    "                                embedding_dim = 25,\n",
    "                                input_size = 100,\n",
    "                                hidden_size1 = 100,\n",
    "                                embedding = text_field.vocab.vectors\n",
    "                                )\n",
    "          optimizer =  optimizer_name(net.parameters(),lr = lr)\n",
    "          train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer)\n",
    "          loss_accVSepochs(train_loss,train_acc,val_loss,val_acc,epochs)\n",
    "          print(\"\\n\\nTEST RESULTS:\")\n",
    "          test(net,test_loader,loss_function)\n",
    "          print(\"----------------------------------------------\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Q_iU0icsoXZd"
   },
   "source": [
    "## __Loss__ and __Accuracy__ VS __Epochs__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "V6QK0n65oaPH"
   },
   "outputs": [],
   "source": [
    "plt.style.use('seaborn-whitegrid') # plot style\n",
    "\n",
    "def loss_accVSepochs(train_loss,train_acc,val_loss,val_acc,epochs):\n",
    "  # plt.subplots_adjust( wspace=0.9, hspace=0.9)\n",
    "  fig, (ax1, ax2) = plt.subplots(1,2,figsize=(10,4))\n",
    "\n",
    "  ax1.plot(range(epochs),train_loss,'bo-' ,label = 'Training loss')\n",
    "  ax1.plot(range(epochs),val_loss ,'ro-',label = 'Validation loss')\n",
    "  ax1.set_ylabel('Loss')\n",
    "  ax1.set_xlabel('Epoch')\n",
    "  ax1.set_title('Loss VS Epoch',y = 1.03,fontsize=15)\n",
    "  ax1.set_xlim([-1,epochs])\n",
    "  ax1.legend(loc=\"best\")\n",
    "\n",
    "  ax2.plot(range(epochs),train_acc,'bo-' ,label = 'Training accuracy')\n",
    "  ax2.plot(range(epochs),val_acc ,'ro-',label = 'Validation accuracy')\n",
    "  ax2.set_ylabel('Accuracy')\n",
    "  ax2.set_xlabel('Epoch')\n",
    "  ax2.set_title('Accuracy VS Epoch',y = 1.03,fontsize=15)\n",
    "  ax2.set_xlim([-1,epochs])\n",
    "  ax2.legend(loc=\"best\")\n",
    "\n",
    "  plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "RP0yiPUZogKG"
   },
   "source": [
    "# __Experiment section__\n",
    "\n",
    "I will test some Neural Nets and their behaviour by changing or adding: \n",
    "* Probability in Dropout Layer\n",
    "* Size of clip in gradient clipping\n",
    "* Number of hidden layers/states\n",
    "* Number of stacked RNNs\n",
    "* Type of cells (RNN,LSTM,GRU)\n",
    "* Skip connections\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "Au180ZceUV1S"
   },
   "outputs": [],
   "source": [
    "from argparse import Namespace\n",
    "\n",
    "args = Namespace(\n",
    "    gradient_clip = 2,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "due2sPSBOw56"
   },
   "source": [
    "## Simple biderectional RNN implementation\n",
    "\n",
    "Implemented with multiple cells, one RNN model, one Linear layer and a sigmoid.\n",
    "\n",
    "Here, we'll instantiate the network. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "rVFhM5nEO4rw"
   },
   "outputs": [],
   "source": [
    "class RNN_TweetClassifier(nn.Module):\n",
    "    def __init__(self, cell_type, input_size, output_size, vocab_size, embedding_dim, hidden_size,num_layers, num_classes=1,hidden_dim=200,hidden_size1 = 500,hidden_size2 = 100, embedding=None ,bidirectional=True ,dropout_prob=0.2):\n",
    "        super(RNN_TweetClassifier,self).__init__()\n",
    "\n",
    "\n",
    "        cells = {\n",
    "          \"RNN\"     : nn.RNN,\n",
    "          \"LSTM\"    : nn.LSTM,\n",
    "          \"GRU\"     : nn.GRU\n",
    "        }\n",
    "\n",
    "        self.cell_type = cell_type\n",
    "        self.dropout_prob = dropout_prob\n",
    "        self.input_size, self.output_size, self.hidden_size, self.vocab_size, self.embedding_dim ,self.num_layers = input_size, output_size, hidden_size, vocab_size, embedding_dim, num_layers\n",
    "        self.bidirectional = bidirectional\n",
    "\n",
    "        # self.embedding = embedding  # or better\n",
    "        self.embedding = nn.Embedding.from_pretrained(torch.FloatTensor(embedding),padding_idx=0)\n",
    "        \n",
    "        self.rnn = cells[cell_type](         # Pick the specific model\n",
    "            input_size=embedding_dim,           # Number of features for each time step\n",
    "            hidden_size=hidden_size,         # Rnn hidden embedding representation\n",
    "            num_layers=num_layers,           # Number of layers (if >1 it is a stacked RNN)\n",
    "            bidirectional=bidirectional,     # If true it is a \n",
    "            batch_first=True,                # input & output will have batch size as 1st dimension. e.g. (batch, time_step, input_size)\n",
    "            dropout = dropout_prob if dropout_prob != None else 0,\n",
    "\n",
    "        )        \n",
    "\n",
    "        if dropout_prob != None:\n",
    "          self.dropout = nn.Dropout(p = dropout_prob)\n",
    "\n",
    "        if bidirectional == True:\n",
    "          self.fc1 = nn.Linear(hidden_size*2 , output_size)        \n",
    "        else:\n",
    "          self.fc1 = nn.Linear(hidden_size , output_size)\n",
    "\n",
    "\n",
    "        # self.fc1 = nn.Linear(input_size* embedding_dim,hidden_size1)\n",
    "        # self.relu1 = nn.ReLU()\n",
    "        \n",
    "        # self.fc2 = nn.Linear(50,1)\n",
    "\n",
    "        self.sigmoid = nn.Sigmoid()\n",
    "        \n",
    "    def forward(self, x):\n",
    "        \n",
    "        # print('x shape:', x.shape)\n",
    "        # out = self.embedding[x].reshape(x.size()[0], -1) or better\n",
    "        out = self.embedding(x)\n",
    "        \n",
    "        if self.dropout_prob != None:\n",
    "          out = self.dropout(out)\n",
    "\n",
    "        # print('out shape:', out.shape) \n",
    "        if self.cell_type == 'LSTM':\n",
    "            r_out, (last_hidden_state, last_cell_state) = self.rnn(out.detach()) \n",
    "        else:\n",
    "            r_out, last_hidden_state = self.rnn(out.detach())\n",
    "\n",
    "\n",
    "        # print(\"r_out\",r_out.shape)\n",
    "        \n",
    "        r_out = r_out[:, -1, :]\n",
    "        \n",
    "\n",
    "        out = self.fc1(r_out) \n",
    "\n",
    "\n",
    "        \n",
    "        # out = self.fc2(out)\n",
    "\n",
    "        out = self.sigmoid(out)\n",
    "\n",
    "        return out"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "-TaryFAuT-C1"
   },
   "source": [
    "### Running this network with 2 layers and __RNN__ cell"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "h2mbETQgOwog",
    "outputId": "1b276d0d-c1a2-4c22-8224-28ac24e2b4fd"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 001 | Train_Loss: 0.70194 | Train_Acc: 53.685 | Val_Loss: 0.697 | Val_Acc: 56.038\n",
      "Epoch 002 | Train_Loss: 0.69984 | Train_Acc: 53.552 | Val_Loss: 0.690 | Val_Acc: 56.018\n",
      "Epoch 003 | Train_Loss: 0.70546 | Train_Acc: 53.504 | Val_Loss: 0.745 | Val_Acc: 56.038\n",
      "Epoch 004 | Train_Loss: 0.70390 | Train_Acc: 53.530 | Val_Loss: 0.686 | Val_Acc: 56.057\n",
      "Epoch 005 | Train_Loss: 0.70899 | Train_Acc: 53.274 | Val_Loss: 0.688 | Val_Acc: 56.077\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.704026 \n",
      "Train mean acc: 53.509119\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.687730 \n",
      "Validation mean acc: 56.077009\n"
     ]
    }
   ],
   "source": [
    "net = RNN_TweetClassifier(\n",
    "    vocab_size = len(text_field.vocab),\n",
    "    embedding_dim = glove_length,\n",
    "    input_size = vector_length,\n",
    "    hidden_size = 128,\n",
    "    output_size = 1,\n",
    "    num_layers = 2,\n",
    "    bidirectional = True,\n",
    "    cell_type = 'RNN',\n",
    "    embedding = text_field.vocab.vectors,\n",
    "    dropout_prob = None   # without dropout\n",
    ")\n",
    "net\n",
    "learning_rate = 0.01\n",
    "epochs = 5\n",
    "loss_function = nn.BCELoss()\n",
    "optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "# Without gradient clipping\n",
    "args.gradient_clip = None\n",
    "\n",
    "train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "8BvKHn__hipL"
   },
   "source": [
    "**Remarks:** \n",
    "\n",
    "From this execution I can see that RNN without gradient clipping and dropout probability has a very bad performance. So bad that the algorithm learns very slow."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "n5xDvz3qUogG"
   },
   "source": [
    "### Running this network with 2 layers and __LSTM__ cell"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "iBRLSVEyU2BE",
    "outputId": "d97cbec6-f4c6-4bb3-bc39-8d6df88cf526"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 001 | Train_Loss: 0.69344 | Train_Acc: 50.190 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69337 | Train_Acc: 50.105 | Val_Loss: 0.693 | Val_Acc: 49.060\n",
      "Epoch 003 | Train_Loss: 0.69330 | Train_Acc: 49.954 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 004 | Train_Loss: 0.69323 | Train_Acc: 49.834 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 005 | Train_Loss: 0.69321 | Train_Acc: 50.481 | Val_Loss: 0.694 | Val_Acc: 49.060\n",
      "Epoch 006 | Train_Loss: 0.69330 | Train_Acc: 49.736 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 007 | Train_Loss: 0.69332 | Train_Acc: 49.972 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 008 | Train_Loss: 0.69327 | Train_Acc: 50.207 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 009 | Train_Loss: 0.69325 | Train_Acc: 50.036 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 010 | Train_Loss: 0.69322 | Train_Acc: 50.161 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.693290 \n",
      "Train mean acc: 50.067658\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.693000 \n",
      "Validation mean acc: 50.940290\n"
     ]
    }
   ],
   "source": [
    "net = RNN_TweetClassifier(\n",
    "    vocab_size = len(text_field.vocab),\n",
    "    embedding_dim = glove_length,\n",
    "    input_size = vector_length,\n",
    "    hidden_size = 128,\n",
    "    output_size = 1,\n",
    "    num_layers = 2,\n",
    "    bidirectional = True,\n",
    "    cell_type = 'LSTM',\n",
    "    embedding = text_field.vocab.vectors,\n",
    "    dropout_prob = None\n",
    ")\n",
    "net\n",
    "learning_rate = 0.01\n",
    "epochs = 10\n",
    "loss_function = nn.BCELoss()\n",
    "optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "# Without gradient clipping\n",
    "args.gradient_clip = None\n",
    "\n",
    "train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "tQaXJ8aBh8Th"
   },
   "source": [
    "**Remarks:** \n",
    "\n",
    "I expected better performance from before as LSTM stores a more information. In contrast I can see that these hyperparameters and LSTM have even worse performance from before. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "weiN7YpPU6_I"
   },
   "source": [
    "### Running this network with 2 layers and __GRU__ cell"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "jaIMGMBFU8gi",
    "outputId": "a2ad2755-5910-4bf9-9a84-24e4f00ddb40"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 001 | Train_Loss: 0.69433 | Train_Acc: 50.041 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69346 | Train_Acc: 49.786 | Val_Loss: 0.693 | Val_Acc: 49.060\n",
      "Epoch 003 | Train_Loss: 0.68952 | Train_Acc: 52.863 | Val_Loss: 0.623 | Val_Acc: 65.608\n",
      "Epoch 004 | Train_Loss: 0.59378 | Train_Acc: 67.976 | Val_Loss: 0.568 | Val_Acc: 69.422\n",
      "Epoch 005 | Train_Loss: 0.56652 | Train_Acc: 70.278 | Val_Loss: 0.559 | Val_Acc: 71.161\n",
      "Epoch 006 | Train_Loss: 0.55468 | Train_Acc: 71.409 | Val_Loss: 0.561 | Val_Acc: 70.385\n",
      "Epoch 007 | Train_Loss: 0.54418 | Train_Acc: 72.234 | Val_Loss: 0.549 | Val_Acc: 72.768\n",
      "Epoch 008 | Train_Loss: 0.53040 | Train_Acc: 73.309 | Val_Loss: 0.540 | Val_Acc: 72.651\n",
      "Epoch 009 | Train_Loss: 0.53106 | Train_Acc: 73.137 | Val_Loss: 0.556 | Val_Acc: 71.844\n",
      "Epoch 010 | Train_Loss: 0.52457 | Train_Acc: 73.703 | Val_Loss: 0.578 | Val_Acc: 71.010\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.592250 \n",
      "Train mean acc: 65.473559\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.578484 \n",
      "Validation mean acc: 71.010045\n"
     ]
    }
   ],
   "source": [
    "net = RNN_TweetClassifier(\n",
    "    vocab_size = len(text_field.vocab),\n",
    "    embedding_dim = glove_length,\n",
    "    input_size = vector_length,\n",
    "    hidden_size = 128,\n",
    "    output_size = 1,\n",
    "    num_layers = 2,\n",
    "    bidirectional = True,\n",
    "    cell_type = 'GRU',\n",
    "    embedding = text_field.vocab.vectors,\n",
    "    dropout_prob = None\n",
    ")\n",
    "net\n",
    "learning_rate = 0.01\n",
    "epochs = 10\n",
    "loss_function = nn.BCELoss()\n",
    "optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "# Without gradient clipping\n",
    "args.gradient_clip = None\n",
    "\n",
    "train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "KasfIsVmicQ8"
   },
   "source": [
    "**Remarks:** This is the best execution so far. GRU is very promissing and also mush quicker from RNN and LSTM  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "7LgEvYMnXRQL"
   },
   "source": [
    "__Remarks__ so far:\n",
    "\n",
    "* I tried different cell types in a simple RNN and the training and validation loss-accuracy aren't so good, since my previous projects (CNN,LogisticRegression) had better results. Especially, my first two tries aren't good at all, since the model doesn't learn much. I expected better performance of LSTM and GRU because these models store more gradients and states. \n",
    "\n",
    "* Also,I tried different learning rates, since every model had a serious problem in learning if I had a small or big learning rate. For this reason I inserted a scheduler, as you proposed, to fix dynamically the learning rates (ReduceLROnPlateau).\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "rHzHr-VvWFoB"
   },
   "source": [
    "## Adding __Gradient clipping__  \n",
    "\n",
    "In this section I will enable gradient clipping in my test function and I will try multiple values for the size of clip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "_XrW1TMRWFMY",
    "outputId": "dcd04d61-49bb-43c7-f3e2-bc1bb77575ae"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "----------------------   RNN  |  5   ---------------------- \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69414 | Train_Acc: 49.845 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69341 | Train_Acc: 49.651 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.68116 | Train_Acc: 55.483 | Val_Loss: 0.621 | Val_Acc: 65.100\n",
      "Epoch 004 | Train_Loss: 0.58957 | Train_Acc: 68.137 | Val_Loss: 0.571 | Val_Acc: 69.568\n",
      "Epoch 005 | Train_Loss: 0.54125 | Train_Acc: 72.589 | Val_Loss: 0.533 | Val_Acc: 73.354\n",
      "Epoch 006 | Train_Loss: 0.51173 | Train_Acc: 74.715 | Val_Loss: 0.519 | Val_Acc: 74.741\n",
      "Epoch 007 | Train_Loss: 0.49735 | Train_Acc: 75.549 | Val_Loss: 0.531 | Val_Acc: 74.506\n",
      "Epoch 008 | Train_Loss: 0.48255 | Train_Acc: 76.597 | Val_Loss: 0.525 | Val_Acc: 73.823\n",
      "Epoch 009 | Train_Loss: 0.48246 | Train_Acc: 76.462 | Val_Loss: 0.534 | Val_Acc: 73.237\n",
      "Epoch 010 | Train_Loss: 0.48528 | Train_Acc: 76.425 | Val_Loss: 0.520 | Val_Acc: 74.643\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.565890 \n",
      "Train mean acc: 67.545477\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.520359 \n",
      "Validation mean acc: 74.642857\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "----------------------   LSTM  |  5   ---------------------- \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69385 | Train_Acc: 49.742 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69330 | Train_Acc: 50.064 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.67675 | Train_Acc: 55.531 | Val_Loss: 0.662 | Val_Acc: 60.159\n",
      "Epoch 004 | Train_Loss: 0.58049 | Train_Acc: 69.155 | Val_Loss: 0.579 | Val_Acc: 69.227\n",
      "Epoch 005 | Train_Loss: 0.53958 | Train_Acc: 72.649 | Val_Loss: 0.561 | Val_Acc: 70.745\n",
      "Epoch 006 | Train_Loss: 0.52357 | Train_Acc: 73.810 | Val_Loss: 0.547 | Val_Acc: 71.864\n",
      "Epoch 007 | Train_Loss: 0.51210 | Train_Acc: 74.692 | Val_Loss: 0.546 | Val_Acc: 71.682\n",
      "Epoch 008 | Train_Loss: 0.50336 | Train_Acc: 75.319 | Val_Loss: 0.539 | Val_Acc: 72.508\n",
      "Epoch 009 | Train_Loss: 0.49214 | Train_Acc: 76.166 | Val_Loss: 0.560 | Val_Acc: 71.532\n",
      "Epoch 010 | Train_Loss: 0.48319 | Train_Acc: 76.534 | Val_Loss: 0.555 | Val_Acc: 71.155\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.569834 \n",
      "Train mean acc: 67.366164\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.554541 \n",
      "Validation mean acc: 71.155134\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "----------------------   GRU  |  5   ---------------------- \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.67330 | Train_Acc: 56.092 | Val_Loss: 0.609 | Val_Acc: 65.784\n",
      "Epoch 002 | Train_Loss: 0.58967 | Train_Acc: 68.152 | Val_Loss: 0.571 | Val_Acc: 70.095\n",
      "Epoch 003 | Train_Loss: 0.55003 | Train_Acc: 71.891 | Val_Loss: 0.551 | Val_Acc: 73.178\n",
      "Epoch 004 | Train_Loss: 0.52666 | Train_Acc: 73.578 | Val_Loss: 0.554 | Val_Acc: 72.020\n",
      "Epoch 005 | Train_Loss: 0.51409 | Train_Acc: 74.689 | Val_Loss: 0.554 | Val_Acc: 72.782\n",
      "Epoch 006 | Train_Loss: 0.50647 | Train_Acc: 75.176 | Val_Loss: 0.537 | Val_Acc: 74.090\n",
      "Epoch 007 | Train_Loss: 0.49943 | Train_Acc: 75.932 | Val_Loss: 0.531 | Val_Acc: 73.108\n",
      "Epoch 008 | Train_Loss: 0.48929 | Train_Acc: 76.310 | Val_Loss: 0.523 | Val_Acc: 74.696\n",
      "Epoch 009 | Train_Loss: 0.48094 | Train_Acc: 76.950 | Val_Loss: 0.595 | Val_Acc: 71.576\n",
      "Epoch 010 | Train_Loss: 0.47993 | Train_Acc: 76.984 | Val_Loss: 0.527 | Val_Acc: 73.836\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.530980 \n",
      "Train mean acc: 72.575479\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.527021 \n",
      "Validation mean acc: 73.836496\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "----------------------   RNN  |  2   ---------------------- \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69444 | Train_Acc: 50.197 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69339 | Train_Acc: 50.202 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.69340 | Train_Acc: 50.110 | Val_Loss: 0.694 | Val_Acc: 50.940\n",
      "Epoch 004 | Train_Loss: 0.70438 | Train_Acc: 53.488 | Val_Loss: 0.693 | Val_Acc: 53.167\n",
      "Epoch 005 | Train_Loss: 0.69161 | Train_Acc: 54.917 | Val_Loss: 0.701 | Val_Acc: 52.882\n",
      "Epoch 006 | Train_Loss: 0.68605 | Train_Acc: 55.421 | Val_Loss: 0.688 | Val_Acc: 55.089\n",
      "Epoch 007 | Train_Loss: 0.68998 | Train_Acc: 55.009 | Val_Loss: 0.691 | Val_Acc: 55.036\n",
      "Epoch 008 | Train_Loss: 0.68618 | Train_Acc: 55.763 | Val_Loss: 0.688 | Val_Acc: 55.876\n",
      "Epoch 009 | Train_Loss: 0.68696 | Train_Acc: 55.515 | Val_Loss: 0.692 | Val_Acc: 55.550\n",
      "Epoch 010 | Train_Loss: 0.68834 | Train_Acc: 55.384 | Val_Loss: 0.692 | Val_Acc: 55.407\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.691472 \n",
      "Train mean acc: 53.600473\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.691811 \n",
      "Validation mean acc: 55.407366\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "----------------------   LSTM  |  2   ---------------------- \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69426 | Train_Acc: 50.088 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69318 | Train_Acc: 50.428 | Val_Loss: 0.694 | Val_Acc: 49.060\n",
      "Epoch 003 | Train_Loss: 0.69918 | Train_Acc: 50.574 | Val_Loss: 0.705 | Val_Acc: 51.624\n",
      "Epoch 004 | Train_Loss: 0.68379 | Train_Acc: 56.022 | Val_Loss: 0.677 | Val_Acc: 58.772\n",
      "Epoch 005 | Train_Loss: 0.68466 | Train_Acc: 56.196 | Val_Loss: 0.684 | Val_Acc: 56.046\n",
      "Epoch 006 | Train_Loss: 0.68007 | Train_Acc: 56.955 | Val_Loss: 0.676 | Val_Acc: 58.616\n",
      "Epoch 007 | Train_Loss: 0.68061 | Train_Acc: 57.110 | Val_Loss: 0.676 | Val_Acc: 59.222\n",
      "Epoch 008 | Train_Loss: 0.67662 | Train_Acc: 58.091 | Val_Loss: 0.675 | Val_Acc: 59.222\n",
      "Epoch 009 | Train_Loss: 0.67668 | Train_Acc: 57.817 | Val_Loss: 0.693 | Val_Acc: 54.573\n",
      "Epoch 010 | Train_Loss: 0.67607 | Train_Acc: 57.819 | Val_Loss: 0.688 | Val_Acc: 55.304\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.684513 \n",
      "Train mean acc: 55.110057\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.688315 \n",
      "Validation mean acc: 55.304129\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "----------------------   GRU  |  2   ---------------------- \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69385 | Train_Acc: 49.943 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69330 | Train_Acc: 49.976 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.62932 | Train_Acc: 62.489 | Val_Loss: 0.594 | Val_Acc: 68.401\n",
      "Epoch 004 | Train_Loss: 0.54652 | Train_Acc: 72.286 | Val_Loss: 0.554 | Val_Acc: 72.040\n",
      "Epoch 005 | Train_Loss: 0.53271 | Train_Acc: 72.901 | Val_Loss: 0.538 | Val_Acc: 72.352\n",
      "Epoch 006 | Train_Loss: 0.50933 | Train_Acc: 74.875 | Val_Loss: 0.539 | Val_Acc: 72.366\n",
      "Epoch 007 | Train_Loss: 0.49826 | Train_Acc: 75.597 | Val_Loss: 0.541 | Val_Acc: 72.874\n",
      "Epoch 008 | Train_Loss: 0.48878 | Train_Acc: 75.924 | Val_Loss: 0.531 | Val_Acc: 74.272\n",
      "Epoch 009 | Train_Loss: 0.47914 | Train_Acc: 76.800 | Val_Loss: 0.523 | Val_Acc: 73.979\n",
      "Epoch 010 | Train_Loss: 0.47935 | Train_Acc: 76.782 | Val_Loss: 0.528 | Val_Acc: 72.586\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.555057 \n",
      "Train mean acc: 68.757205\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.528313 \n",
      "Validation mean acc: 72.586496\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "----------------------   RNN  |  1   ---------------------- \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69434 | Train_Acc: 50.349 | Val_Loss: 0.693 | Val_Acc: 49.060\n",
      "Epoch 002 | Train_Loss: 0.67929 | Train_Acc: 54.390 | Val_Loss: 0.624 | Val_Acc: 64.554\n",
      "Epoch 003 | Train_Loss: 0.57668 | Train_Acc: 69.232 | Val_Loss: 0.577 | Val_Acc: 69.235\n",
      "Epoch 004 | Train_Loss: 0.53461 | Train_Acc: 72.751 | Val_Loss: 0.543 | Val_Acc: 72.860\n",
      "Epoch 005 | Train_Loss: 0.52158 | Train_Acc: 73.932 | Val_Loss: 0.527 | Val_Acc: 72.854\n",
      "Epoch 006 | Train_Loss: 0.51706 | Train_Acc: 74.467 | Val_Loss: 0.542 | Val_Acc: 72.665\n",
      "Epoch 007 | Train_Loss: 0.51091 | Train_Acc: 74.537 | Val_Loss: 0.529 | Val_Acc: 73.036\n",
      "Epoch 008 | Train_Loss: 0.50497 | Train_Acc: 75.046 | Val_Loss: 0.533 | Val_Acc: 73.290\n",
      "Epoch 009 | Train_Loss: 0.51586 | Train_Acc: 74.483 | Val_Loss: 0.532 | Val_Acc: 73.569\n",
      "Epoch 010 | Train_Loss: 0.50473 | Train_Acc: 75.015 | Val_Loss: 0.525 | Val_Acc: 74.116\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.556003 \n",
      "Train mean acc: 69.420194\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.524555 \n",
      "Validation mean acc: 74.115513\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "----------------------   LSTM  |  1   ---------------------- \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69412 | Train_Acc: 50.155 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69334 | Train_Acc: 50.234 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.70306 | Train_Acc: 51.124 | Val_Loss: 0.674 | Val_Acc: 58.298\n",
      "Epoch 004 | Train_Loss: 0.61497 | Train_Acc: 66.236 | Val_Loss: 0.590 | Val_Acc: 68.206\n",
      "Epoch 005 | Train_Loss: 0.56794 | Train_Acc: 70.262 | Val_Loss: 0.562 | Val_Acc: 71.044\n",
      "Epoch 006 | Train_Loss: 0.55274 | Train_Acc: 71.655 | Val_Loss: 0.544 | Val_Acc: 72.377\n",
      "Epoch 007 | Train_Loss: 0.54014 | Train_Acc: 72.617 | Val_Loss: 0.548 | Val_Acc: 72.768\n",
      "Epoch 008 | Train_Loss: 0.52685 | Train_Acc: 73.580 | Val_Loss: 0.535 | Val_Acc: 73.641\n",
      "Epoch 009 | Train_Loss: 0.51626 | Train_Acc: 74.579 | Val_Loss: 0.536 | Val_Acc: 72.762\n",
      "Epoch 010 | Train_Loss: 0.51048 | Train_Acc: 74.921 | Val_Loss: 0.552 | Val_Acc: 72.653\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.591989 \n",
      "Train mean acc: 65.536275\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.551600 \n",
      "Validation mean acc: 72.653460\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "----------------------   GRU  |  1   ---------------------- \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69412 | Train_Acc: 49.992 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69337 | Train_Acc: 50.049 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.64745 | Train_Acc: 61.471 | Val_Loss: 0.625 | Val_Acc: 66.278\n",
      "Epoch 004 | Train_Loss: 0.59465 | Train_Acc: 68.153 | Val_Loss: 0.587 | Val_Acc: 68.811\n",
      "Epoch 005 | Train_Loss: 0.56677 | Train_Acc: 70.168 | Val_Loss: 0.578 | Val_Acc: 68.870\n",
      "Epoch 006 | Train_Loss: 0.55198 | Train_Acc: 71.423 | Val_Loss: 0.558 | Val_Acc: 71.434\n",
      "Epoch 007 | Train_Loss: 0.54147 | Train_Acc: 72.264 | Val_Loss: 0.547 | Val_Acc: 71.278\n",
      "Epoch 008 | Train_Loss: 0.53138 | Train_Acc: 72.903 | Val_Loss: 0.551 | Val_Acc: 71.479\n",
      "Epoch 009 | Train_Loss: 0.52547 | Train_Acc: 73.310 | Val_Loss: 0.536 | Val_Acc: 73.192\n",
      "Epoch 010 | Train_Loss: 0.51897 | Train_Acc: 74.130 | Val_Loss: 0.538 | Val_Acc: 71.869\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.586564 \n",
      "Train mean acc: 66.386049\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.537657 \n",
      "Validation mean acc: 71.869420\n",
      "\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "cells = ['RNN','LSTM','GRU']\n",
    "clips  = [5,2,1]\n",
    "\n",
    "for clip in clips:\n",
    "\n",
    "  for cell in cells:\n",
    "\n",
    "    print(\"\\n----------------------  \",cell,\" | \",clip,\"  ---------------------- \\n\")\n",
    "    net = RNN_TweetClassifier(\n",
    "        vocab_size = len(text_field.vocab),\n",
    "        embedding_dim = glove_length,\n",
    "        input_size = vector_length,\n",
    "        hidden_size = 128,\n",
    "        output_size = 1,\n",
    "        num_layers = 2,\n",
    "        bidirectional = True,\n",
    "        cell_type = 'GRU',\n",
    "        embedding = text_field.vocab.vectors,\n",
    "        dropout_prob = None\n",
    "    )\n",
    "    net\n",
    "    learning_rate = 0.01\n",
    "    epochs = 10\n",
    "    loss_function = nn.BCELoss()\n",
    "    optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "    lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "    # Without gradient clipping\n",
    "    args.gradient_clip = clip\n",
    "\n",
    "    train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)\n",
    "    print(\"\\n\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "a0fhlquLbr96"
   },
   "source": [
    "__Remarks__:\n",
    "\n",
    "* It is obvious that gradient clipping made the network train better overall and as a result I get better results from previous experiment. Again, LSTM and GRU have better results.\n",
    "\n",
    "* As for the size of clip, I can see multiple behaviours, but I can see that clip size 5 has good results and clip size 2 has some not so good results. For this reason I am going to experiment with both to see with more details that behaviour.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "GtiXbAmmWQmD"
   },
   "source": [
    "## Adding __Dropout propability__\n",
    "\n",
    "In this section I will add a dropout layer after the embedding and also a dropout probability in RNN layers.\n",
    "\n",
    "I will also keep the best gradient clipping size from previous expriment"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "TJQCwpysWbBs",
    "outputId": "692fa137-8109-4e17-f3d3-478198c8fbb5"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "---------   LSTM  |  0.9   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.70686 | Train_Acc: 50.063 | Val_Loss: 0.706 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.70897 | Train_Acc: 49.900 | Val_Loss: 0.692 | Val_Acc: 51.551\n",
      "Epoch 003 | Train_Loss: 0.70925 | Train_Acc: 49.994 | Val_Loss: 0.697 | Val_Acc: 48.825\n",
      "Epoch 004 | Train_Loss: 0.70958 | Train_Acc: 50.101 | Val_Loss: 0.692 | Val_Acc: 52.210\n",
      "Epoch 005 | Train_Loss: 0.70856 | Train_Acc: 50.211 | Val_Loss: 0.756 | Val_Acc: 50.940\n",
      "Epoch 006 | Train_Loss: 0.70740 | Train_Acc: 50.109 | Val_Loss: 0.696 | Val_Acc: 51.038\n",
      "Epoch 007 | Train_Loss: 0.71016 | Train_Acc: 49.665 | Val_Loss: 0.693 | Val_Acc: 51.415\n",
      "Epoch 008 | Train_Loss: 0.70706 | Train_Acc: 50.109 | Val_Loss: 0.691 | Val_Acc: 51.800\n",
      "Epoch 009 | Train_Loss: 0.70729 | Train_Acc: 50.775 | Val_Loss: 0.689 | Val_Acc: 54.015\n",
      "Epoch 010 | Train_Loss: 0.71353 | Train_Acc: 50.256 | Val_Loss: 0.711 | Val_Acc: 49.060\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.708866 \n",
      "Train mean acc: 50.118327\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.711419 \n",
      "Validation mean acc: 49.059710\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  0.9   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.70298 | Train_Acc: 50.205 | Val_Loss: 0.692 | Val_Acc: 51.448\n",
      "Epoch 002 | Train_Loss: 0.71255 | Train_Acc: 50.491 | Val_Loss: 0.693 | Val_Acc: 48.795\n",
      "Epoch 003 | Train_Loss: 0.70968 | Train_Acc: 50.305 | Val_Loss: 0.692 | Val_Acc: 51.468\n",
      "Epoch 004 | Train_Loss: 0.71042 | Train_Acc: 49.526 | Val_Loss: 0.706 | Val_Acc: 50.940\n",
      "Epoch 005 | Train_Loss: 0.70663 | Train_Acc: 50.073 | Val_Loss: 0.700 | Val_Acc: 49.079\n",
      "Epoch 006 | Train_Loss: 0.70724 | Train_Acc: 50.044 | Val_Loss: 0.696 | Val_Acc: 48.518\n",
      "Epoch 007 | Train_Loss: 0.70763 | Train_Acc: 50.185 | Val_Loss: 0.709 | Val_Acc: 50.940\n",
      "Epoch 008 | Train_Loss: 0.70520 | Train_Acc: 50.091 | Val_Loss: 0.708 | Val_Acc: 50.940\n",
      "Epoch 009 | Train_Loss: 0.70717 | Train_Acc: 50.193 | Val_Loss: 0.712 | Val_Acc: 50.940\n",
      "Epoch 010 | Train_Loss: 0.70819 | Train_Acc: 50.405 | Val_Loss: 0.693 | Val_Acc: 51.800\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.707770 \n",
      "Train mean acc: 50.151724\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.692527 \n",
      "Validation mean acc: 51.799665\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   LSTM  |  0.5   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69446 | Train_Acc: 50.101 | Val_Loss: 0.694 | Val_Acc: 51.136\n",
      "Epoch 002 | Train_Loss: 0.70399 | Train_Acc: 50.701 | Val_Loss: 0.690 | Val_Acc: 55.628\n",
      "Epoch 003 | Train_Loss: 0.69795 | Train_Acc: 51.165 | Val_Loss: 0.691 | Val_Acc: 51.390\n",
      "Epoch 004 | Train_Loss: 0.69745 | Train_Acc: 51.430 | Val_Loss: 0.690 | Val_Acc: 51.370\n",
      "Epoch 005 | Train_Loss: 0.69990 | Train_Acc: 51.317 | Val_Loss: 0.689 | Val_Acc: 52.927\n",
      "Epoch 006 | Train_Loss: 0.69978 | Train_Acc: 51.027 | Val_Loss: 0.690 | Val_Acc: 54.900\n",
      "Epoch 007 | Train_Loss: 0.69894 | Train_Acc: 51.023 | Val_Loss: 0.693 | Val_Acc: 49.724\n",
      "Epoch 008 | Train_Loss: 0.69691 | Train_Acc: 51.721 | Val_Loss: 0.690 | Val_Acc: 52.316\n",
      "Epoch 009 | Train_Loss: 0.70107 | Train_Acc: 51.089 | Val_Loss: 0.692 | Val_Acc: 51.018\n",
      "Epoch 010 | Train_Loss: 0.69822 | Train_Acc: 51.655 | Val_Loss: 0.702 | Val_Acc: 49.255\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.698869 \n",
      "Train mean acc: 51.122853\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.701609 \n",
      "Validation mean acc: 49.255022\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  0.5   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.68506 | Train_Acc: 54.098 | Val_Loss: 0.621 | Val_Acc: 64.152\n",
      "Epoch 002 | Train_Loss: 0.63863 | Train_Acc: 63.695 | Val_Loss: 0.622 | Val_Acc: 65.896\n",
      "Epoch 003 | Train_Loss: 0.63624 | Train_Acc: 63.556 | Val_Loss: 0.660 | Val_Acc: 61.892\n",
      "Epoch 004 | Train_Loss: 0.64072 | Train_Acc: 63.120 | Val_Loss: 0.612 | Val_Acc: 65.544\n",
      "Epoch 005 | Train_Loss: 0.63804 | Train_Acc: 63.439 | Val_Loss: 0.647 | Val_Acc: 62.757\n",
      "Epoch 006 | Train_Loss: 0.64050 | Train_Acc: 63.294 | Val_Loss: 0.613 | Val_Acc: 65.935\n",
      "Epoch 007 | Train_Loss: 0.64394 | Train_Acc: 62.639 | Val_Loss: 0.617 | Val_Acc: 65.558\n",
      "Epoch 008 | Train_Loss: 0.63908 | Train_Acc: 63.160 | Val_Loss: 0.618 | Val_Acc: 66.219\n",
      "Epoch 009 | Train_Loss: 0.64653 | Train_Acc: 61.984 | Val_Loss: 0.623 | Val_Acc: 65.446\n",
      "Epoch 010 | Train_Loss: 0.64225 | Train_Acc: 63.222 | Val_Loss: 0.629 | Val_Acc: 63.401\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.645099 \n",
      "Train mean acc: 62.220714\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.628823 \n",
      "Validation mean acc: 63.401228\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   LSTM  |  0.2   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.66859 | Train_Acc: 57.180 | Val_Loss: 0.604 | Val_Acc: 67.885\n",
      "Epoch 002 | Train_Loss: 0.61080 | Train_Acc: 66.096 | Val_Loss: 0.577 | Val_Acc: 69.461\n",
      "Epoch 003 | Train_Loss: 0.58816 | Train_Acc: 68.422 | Val_Loss: 0.578 | Val_Acc: 70.086\n",
      "Epoch 004 | Train_Loss: 0.57699 | Train_Acc: 69.231 | Val_Loss: 0.547 | Val_Acc: 71.844\n",
      "Epoch 005 | Train_Loss: 0.56573 | Train_Acc: 70.336 | Val_Loss: 0.565 | Val_Acc: 71.473\n",
      "Epoch 006 | Train_Loss: 0.55729 | Train_Acc: 71.086 | Val_Loss: 0.536 | Val_Acc: 72.762\n",
      "Epoch 007 | Train_Loss: 0.55522 | Train_Acc: 71.485 | Val_Loss: 0.531 | Val_Acc: 73.510\n",
      "Epoch 008 | Train_Loss: 0.54905 | Train_Acc: 71.747 | Val_Loss: 0.533 | Val_Acc: 73.959\n",
      "Epoch 009 | Train_Loss: 0.54273 | Train_Acc: 72.191 | Val_Loss: 0.542 | Val_Acc: 71.844\n",
      "Epoch 010 | Train_Loss: 0.53853 | Train_Acc: 72.645 | Val_Loss: 0.524 | Val_Acc: 73.915\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.575309 \n",
      "Train mean acc: 69.041950\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.524203 \n",
      "Validation mean acc: 73.914621\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  0.2   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69415 | Train_Acc: 49.757 | Val_Loss: 0.693 | Val_Acc: 49.060\n",
      "Epoch 002 | Train_Loss: 0.68842 | Train_Acc: 54.045 | Val_Loss: 0.673 | Val_Acc: 59.289\n",
      "Epoch 003 | Train_Loss: 0.65278 | Train_Acc: 61.794 | Val_Loss: 0.637 | Val_Acc: 63.161\n",
      "Epoch 004 | Train_Loss: 0.62608 | Train_Acc: 64.535 | Val_Loss: 0.599 | Val_Acc: 66.833\n",
      "Epoch 005 | Train_Loss: 0.61774 | Train_Acc: 65.757 | Val_Loss: 0.595 | Val_Acc: 68.357\n",
      "Epoch 006 | Train_Loss: 0.60243 | Train_Acc: 67.265 | Val_Loss: 0.574 | Val_Acc: 69.300\n",
      "Epoch 007 | Train_Loss: 0.58936 | Train_Acc: 68.296 | Val_Loss: 0.593 | Val_Acc: 67.991\n",
      "Epoch 008 | Train_Loss: 0.58399 | Train_Acc: 68.794 | Val_Loss: 0.557 | Val_Acc: 70.569\n",
      "Epoch 009 | Train_Loss: 0.57911 | Train_Acc: 69.458 | Val_Loss: 0.568 | Val_Acc: 69.241\n",
      "Epoch 010 | Train_Loss: 0.57154 | Train_Acc: 70.223 | Val_Loss: 0.539 | Val_Acc: 72.567\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.620561 \n",
      "Train mean acc: 63.992412\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.539323 \n",
      "Validation mean acc: 72.566964\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   LSTM  |  0.1   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.65328 | Train_Acc: 59.029 | Val_Loss: 0.586 | Val_Acc: 69.520\n",
      "Epoch 002 | Train_Loss: 0.58748 | Train_Acc: 68.450 | Val_Loss: 0.590 | Val_Acc: 69.051\n",
      "Epoch 003 | Train_Loss: 0.55541 | Train_Acc: 71.268 | Val_Loss: 0.532 | Val_Acc: 72.737\n",
      "Epoch 004 | Train_Loss: 0.53987 | Train_Acc: 72.263 | Val_Loss: 0.535 | Val_Acc: 72.366\n",
      "Epoch 005 | Train_Loss: 0.52927 | Train_Acc: 73.483 | Val_Loss: 0.537 | Val_Acc: 72.302\n",
      "Epoch 006 | Train_Loss: 0.51422 | Train_Acc: 74.306 | Val_Loss: 0.521 | Val_Acc: 74.501\n",
      "Epoch 007 | Train_Loss: 0.50772 | Train_Acc: 75.015 | Val_Loss: 0.513 | Val_Acc: 74.422\n",
      "Epoch 008 | Train_Loss: 0.50190 | Train_Acc: 75.240 | Val_Loss: 0.527 | Val_Acc: 72.673\n",
      "Epoch 009 | Train_Loss: 0.49407 | Train_Acc: 75.851 | Val_Loss: 0.523 | Val_Acc: 74.442\n",
      "Epoch 010 | Train_Loss: 0.48889 | Train_Acc: 76.089 | Val_Loss: 0.519 | Val_Acc: 74.554\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.537211 \n",
      "Train mean acc: 72.099458\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.519293 \n",
      "Validation mean acc: 74.553571\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  0.1   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.64068 | Train_Acc: 61.404 | Val_Loss: 0.622 | Val_Acc: 66.967\n",
      "Epoch 002 | Train_Loss: 0.57913 | Train_Acc: 69.249 | Val_Loss: 0.559 | Val_Acc: 69.104\n",
      "Epoch 003 | Train_Loss: 0.56333 | Train_Acc: 70.538 | Val_Loss: 0.560 | Val_Acc: 70.432\n",
      "Epoch 004 | Train_Loss: 0.56148 | Train_Acc: 70.657 | Val_Loss: 0.546 | Val_Acc: 71.057\n",
      "Epoch 005 | Train_Loss: 0.54700 | Train_Acc: 71.603 | Val_Loss: 0.546 | Val_Acc: 72.143\n",
      "Epoch 006 | Train_Loss: 0.54384 | Train_Acc: 71.920 | Val_Loss: 0.536 | Val_Acc: 72.444\n",
      "Epoch 007 | Train_Loss: 0.53686 | Train_Acc: 72.743 | Val_Loss: 0.543 | Val_Acc: 71.800\n",
      "Epoch 008 | Train_Loss: 0.53422 | Train_Acc: 73.025 | Val_Loss: 0.530 | Val_Acc: 72.835\n",
      "Epoch 009 | Train_Loss: 0.53114 | Train_Acc: 73.221 | Val_Loss: 0.533 | Val_Acc: 72.522\n",
      "Epoch 010 | Train_Loss: 0.52872 | Train_Acc: 73.243 | Val_Loss: 0.527 | Val_Acc: 74.037\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.556640 \n",
      "Train mean acc: 70.760450\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.526753 \n",
      "Validation mean acc: 74.037388\n",
      "\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "cells = ['LSTM','GRU']\n",
    "probabilities = [0.9,0.5,0.2,0.1]\n",
    "\n",
    "for prob in probabilities:\n",
    "  for cell in cells:\n",
    "\n",
    "    print(\"\\n---------  \",cell,\" | \",prob,\"  ------------ \\n\")\n",
    "    net = RNN_TweetClassifier(\n",
    "        vocab_size = len(text_field.vocab),\n",
    "        embedding_dim = glove_length,\n",
    "        input_size = vector_length,\n",
    "        hidden_size = 128,\n",
    "        output_size = 1,\n",
    "        num_layers = 2,\n",
    "        bidirectional = True,\n",
    "        cell_type = 'GRU',\n",
    "        embedding = text_field.vocab.vectors,\n",
    "        dropout_prob = prob\n",
    "    )\n",
    "    net\n",
    "    learning_rate = 0.01\n",
    "    epochs = 10\n",
    "    loss_function = nn.BCELoss()\n",
    "    optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "    lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "    args.gradient_clip = 5\n",
    "\n",
    "    train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)\n",
    "    print(\"\\n\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "l0TJM7sFjtTt"
   },
   "source": [
    "**Remarks:** \n",
    "In this experiment I keeped only LSTM and GRU as the quickest and most efficient algos.\n",
    "\n",
    "- 0.9 has the performance I expected, which is the worst in this test as this dropout is very big.\n",
    "\n",
    "- As the dropout size decreases the scores and the convergence of the algorithm get better. I can see that 0.1, 0.2 have more or less the same results.\n",
    "\n",
    "- 0.5 is stil big as I can see from the results."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "CzqOPqaNcarQ"
   },
   "source": [
    "## #Number of hidden layers in the simple RNN model\n",
    "\n",
    "I will change the hidden states to see whats the difference in results."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "AIi49g_RcVne",
    "outputId": "42f1fbed-6a46-4d57-8e4e-145b6695a2e8"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "---------   RNN  |  64   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69375 | Train_Acc: 49.734 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69357 | Train_Acc: 50.052 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.69340 | Train_Acc: 50.010 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 004 | Train_Loss: 0.67303 | Train_Acc: 55.851 | Val_Loss: 0.611 | Val_Acc: 66.496\n",
      "Epoch 005 | Train_Loss: 0.61984 | Train_Acc: 65.871 | Val_Loss: 0.596 | Val_Acc: 68.993\n",
      "Epoch 006 | Train_Loss: 0.60728 | Train_Acc: 66.584 | Val_Loss: 0.600 | Val_Acc: 67.659\n",
      "Epoch 007 | Train_Loss: 0.60380 | Train_Acc: 67.155 | Val_Loss: 0.591 | Val_Acc: 66.364\n",
      "Epoch 008 | Train_Loss: 0.59731 | Train_Acc: 67.602 | Val_Loss: 0.585 | Val_Acc: 68.284\n",
      "Epoch 009 | Train_Loss: 0.59397 | Train_Acc: 67.878 | Val_Loss: 0.578 | Val_Acc: 69.294\n",
      "Epoch 010 | Train_Loss: 0.58547 | Train_Acc: 68.780 | Val_Loss: 0.573 | Val_Acc: 69.774\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.636142 \n",
      "Train mean acc: 60.951761\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.573212 \n",
      "Validation mean acc: 69.773996\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   LSTM  |  64   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69369 | Train_Acc: 50.036 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69337 | Train_Acc: 50.410 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.69224 | Train_Acc: 50.814 | Val_Loss: 0.635 | Val_Acc: 65.399\n",
      "Epoch 004 | Train_Loss: 0.61479 | Train_Acc: 66.057 | Val_Loss: 0.576 | Val_Acc: 69.007\n",
      "Epoch 005 | Train_Loss: 0.59272 | Train_Acc: 68.112 | Val_Loss: 0.563 | Val_Acc: 70.901\n",
      "Epoch 006 | Train_Loss: 0.58386 | Train_Acc: 68.785 | Val_Loss: 0.571 | Val_Acc: 70.165\n",
      "Epoch 007 | Train_Loss: 0.57680 | Train_Acc: 69.204 | Val_Loss: 0.548 | Val_Acc: 72.377\n",
      "Epoch 008 | Train_Loss: 0.56557 | Train_Acc: 70.165 | Val_Loss: 0.532 | Val_Acc: 72.098\n",
      "Epoch 009 | Train_Loss: 0.56207 | Train_Acc: 70.577 | Val_Loss: 0.535 | Val_Acc: 73.002\n",
      "Epoch 010 | Train_Loss: 0.55719 | Train_Acc: 70.935 | Val_Loss: 0.538 | Val_Acc: 72.645\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.613228 \n",
      "Train mean acc: 63.509535\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.537826 \n",
      "Validation mean acc: 72.645089\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  64   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69378 | Train_Acc: 50.413 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.69448 | Train_Acc: 50.197 | Val_Loss: 0.695 | Val_Acc: 49.339\n",
      "Epoch 003 | Train_Loss: 0.64689 | Train_Acc: 61.358 | Val_Loss: 0.582 | Val_Acc: 68.493\n",
      "Epoch 004 | Train_Loss: 0.59049 | Train_Acc: 68.446 | Val_Loss: 0.562 | Val_Acc: 70.257\n",
      "Epoch 005 | Train_Loss: 0.57422 | Train_Acc: 69.771 | Val_Loss: 0.544 | Val_Acc: 72.645\n",
      "Epoch 006 | Train_Loss: 0.55827 | Train_Acc: 71.078 | Val_Loss: 0.538 | Val_Acc: 72.879\n",
      "Epoch 007 | Train_Loss: 0.54977 | Train_Acc: 71.535 | Val_Loss: 0.533 | Val_Acc: 72.893\n",
      "Epoch 008 | Train_Loss: 0.54713 | Train_Acc: 71.987 | Val_Loss: 0.565 | Val_Acc: 70.438\n",
      "Epoch 009 | Train_Loss: 0.54360 | Train_Acc: 72.222 | Val_Loss: 0.527 | Val_Acc: 72.821\n",
      "Epoch 010 | Train_Loss: 0.53851 | Train_Acc: 72.547 | Val_Loss: 0.525 | Val_Acc: 74.057\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.593715 \n",
      "Train mean acc: 65.955305\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.524505 \n",
      "Validation mean acc: 74.056920\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   RNN  |  128   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.70262 | Train_Acc: 49.989 | Val_Loss: 0.711 | Val_Acc: 49.372\n",
      "Epoch 002 | Train_Loss: 0.69797 | Train_Acc: 50.745 | Val_Loss: 0.709 | Val_Acc: 49.802\n",
      "Epoch 003 | Train_Loss: 0.70163 | Train_Acc: 50.528 | Val_Loss: 0.694 | Val_Acc: 49.713\n",
      "Epoch 004 | Train_Loss: 0.69876 | Train_Acc: 50.875 | Val_Loss: 0.721 | Val_Acc: 51.057\n",
      "Epoch 005 | Train_Loss: 0.69973 | Train_Acc: 50.749 | Val_Loss: 0.696 | Val_Acc: 51.150\n",
      "Epoch 006 | Train_Loss: 0.69939 | Train_Acc: 50.588 | Val_Loss: 0.696 | Val_Acc: 52.073\n",
      "Epoch 007 | Train_Loss: 0.69968 | Train_Acc: 50.790 | Val_Loss: 0.697 | Val_Acc: 51.038\n",
      "Epoch 008 | Train_Loss: 0.69917 | Train_Acc: 50.987 | Val_Loss: 0.696 | Val_Acc: 52.268\n",
      "Epoch 009 | Train_Loss: 0.69898 | Train_Acc: 50.970 | Val_Loss: 0.708 | Val_Acc: 51.560\n",
      "Epoch 010 | Train_Loss: 0.70226 | Train_Acc: 51.031 | Val_Loss: 0.699 | Val_Acc: 50.485\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.700019 \n",
      "Train mean acc: 50.725156\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.698785 \n",
      "Validation mean acc: 50.485491\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   LSTM  |  128   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69439 | Train_Acc: 49.827 | Val_Loss: 0.693 | Val_Acc: 49.060\n",
      "Epoch 002 | Train_Loss: 0.69853 | Train_Acc: 50.423 | Val_Loss: 0.720 | Val_Acc: 49.568\n",
      "Epoch 003 | Train_Loss: 0.70813 | Train_Acc: 51.414 | Val_Loss: 0.692 | Val_Acc: 52.277\n",
      "Epoch 004 | Train_Loss: 0.69906 | Train_Acc: 52.074 | Val_Loss: 0.713 | Val_Acc: 49.886\n",
      "Epoch 005 | Train_Loss: 0.69661 | Train_Acc: 52.478 | Val_Loss: 0.691 | Val_Acc: 51.956\n",
      "Epoch 006 | Train_Loss: 0.70080 | Train_Acc: 52.105 | Val_Loss: 0.693 | Val_Acc: 52.366\n",
      "Epoch 007 | Train_Loss: 0.69613 | Train_Acc: 52.390 | Val_Loss: 0.696 | Val_Acc: 50.564\n",
      "Epoch 008 | Train_Loss: 0.69804 | Train_Acc: 51.892 | Val_Loss: 0.712 | Val_Acc: 49.275\n",
      "Epoch 009 | Train_Loss: 0.69810 | Train_Acc: 52.534 | Val_Loss: 0.690 | Val_Acc: 52.790\n",
      "Epoch 010 | Train_Loss: 0.69759 | Train_Acc: 51.814 | Val_Loss: 0.762 | Val_Acc: 50.940\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.698738 \n",
      "Train mean acc: 51.695038\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.761620 \n",
      "Validation mean acc: 50.940290\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  128   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69633 | Train_Acc: 49.895 | Val_Loss: 0.689 | Val_Acc: 55.519\n",
      "Epoch 002 | Train_Loss: 0.62928 | Train_Acc: 64.332 | Val_Loss: 0.587 | Val_Acc: 68.524\n",
      "Epoch 003 | Train_Loss: 0.60507 | Train_Acc: 66.807 | Val_Loss: 0.583 | Val_Acc: 69.138\n",
      "Epoch 004 | Train_Loss: 0.59493 | Train_Acc: 67.593 | Val_Loss: 0.576 | Val_Acc: 69.143\n",
      "Epoch 005 | Train_Loss: 0.59279 | Train_Acc: 67.929 | Val_Loss: 0.578 | Val_Acc: 69.847\n",
      "Epoch 006 | Train_Loss: 0.58737 | Train_Acc: 68.583 | Val_Loss: 0.569 | Val_Acc: 69.743\n",
      "Epoch 007 | Train_Loss: 0.58989 | Train_Acc: 68.058 | Val_Loss: 0.582 | Val_Acc: 68.694\n",
      "Epoch 008 | Train_Loss: 0.58596 | Train_Acc: 68.491 | Val_Loss: 0.601 | Val_Acc: 67.567\n",
      "Epoch 009 | Train_Loss: 0.58361 | Train_Acc: 68.868 | Val_Loss: 0.561 | Val_Acc: 71.010\n",
      "Epoch 010 | Train_Loss: 0.58611 | Train_Acc: 68.651 | Val_Loss: 0.573 | Val_Acc: 69.467\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.605134 \n",
      "Train mean acc: 65.920760\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.572752 \n",
      "Validation mean acc: 69.467076\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   RNN  |  256   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.70855 | Train_Acc: 51.196 | Val_Loss: 0.690 | Val_Acc: 53.278\n",
      "Epoch 002 | Train_Loss: 0.69789 | Train_Acc: 52.404 | Val_Loss: 0.707 | Val_Acc: 53.239\n",
      "Epoch 003 | Train_Loss: 0.67494 | Train_Acc: 58.155 | Val_Loss: 0.670 | Val_Acc: 58.669\n",
      "Epoch 004 | Train_Loss: 0.65688 | Train_Acc: 61.763 | Val_Loss: 0.682 | Val_Acc: 60.296\n",
      "Epoch 005 | Train_Loss: 0.64318 | Train_Acc: 63.623 | Val_Loss: 0.622 | Val_Acc: 64.858\n",
      "Epoch 006 | Train_Loss: 0.63820 | Train_Acc: 64.834 | Val_Loss: 0.607 | Val_Acc: 67.670\n",
      "Epoch 007 | Train_Loss: 0.63421 | Train_Acc: 64.911 | Val_Loss: 0.609 | Val_Acc: 66.948\n",
      "Epoch 008 | Train_Loss: 0.64142 | Train_Acc: 64.550 | Val_Loss: 0.606 | Val_Acc: 67.299\n",
      "Epoch 009 | Train_Loss: 0.62477 | Train_Acc: 65.982 | Val_Loss: 0.604 | Val_Acc: 67.084\n",
      "Epoch 010 | Train_Loss: 0.62042 | Train_Acc: 66.406 | Val_Loss: 0.594 | Val_Acc: 68.881\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.654046 \n",
      "Train mean acc: 61.382388\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.593961 \n",
      "Validation mean acc: 68.881138\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   LSTM  |  256   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.71005 | Train_Acc: 51.026 | Val_Loss: 0.695 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.70409 | Train_Acc: 53.119 | Val_Loss: 0.701 | Val_Acc: 54.216\n",
      "Epoch 003 | Train_Loss: 0.68245 | Train_Acc: 57.238 | Val_Loss: 0.689 | Val_Acc: 55.349\n",
      "Epoch 004 | Train_Loss: 0.67047 | Train_Acc: 59.642 | Val_Loss: 0.644 | Val_Acc: 63.290\n",
      "Epoch 005 | Train_Loss: 0.64821 | Train_Acc: 62.232 | Val_Loss: 0.625 | Val_Acc: 65.965\n",
      "Epoch 006 | Train_Loss: 0.63114 | Train_Acc: 64.179 | Val_Loss: 0.616 | Val_Acc: 65.511\n",
      "Epoch 007 | Train_Loss: 0.62723 | Train_Acc: 65.024 | Val_Loss: 0.606 | Val_Acc: 65.642\n",
      "Epoch 008 | Train_Loss: 0.61834 | Train_Acc: 65.870 | Val_Loss: 0.627 | Val_Acc: 65.647\n",
      "Epoch 009 | Train_Loss: 0.61182 | Train_Acc: 66.232 | Val_Loss: 0.588 | Val_Acc: 66.306\n",
      "Epoch 010 | Train_Loss: 0.61050 | Train_Acc: 66.255 | Val_Loss: 0.581 | Val_Acc: 69.026\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.651429 \n",
      "Train mean acc: 61.081753\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.581157 \n",
      "Validation mean acc: 69.026228\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  256   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.70377 | Train_Acc: 50.959 | Val_Loss: 0.701 | Val_Acc: 49.275\n",
      "Epoch 002 | Train_Loss: 0.69425 | Train_Acc: 52.981 | Val_Loss: 0.693 | Val_Acc: 53.987\n",
      "Epoch 003 | Train_Loss: 0.68827 | Train_Acc: 55.532 | Val_Loss: 0.681 | Val_Acc: 56.775\n",
      "Epoch 004 | Train_Loss: 0.68550 | Train_Acc: 56.129 | Val_Loss: 0.688 | Val_Acc: 56.970\n",
      "Epoch 005 | Train_Loss: 0.68109 | Train_Acc: 57.408 | Val_Loss: 0.673 | Val_Acc: 58.435\n",
      "Epoch 006 | Train_Loss: 0.67529 | Train_Acc: 58.211 | Val_Loss: 0.703 | Val_Acc: 56.468\n",
      "Epoch 007 | Train_Loss: 0.67184 | Train_Acc: 58.957 | Val_Loss: 0.668 | Val_Acc: 58.454\n",
      "Epoch 008 | Train_Loss: 0.66399 | Train_Acc: 60.116 | Val_Loss: 0.643 | Val_Acc: 62.972\n",
      "Epoch 009 | Train_Loss: 0.65532 | Train_Acc: 61.327 | Val_Loss: 0.640 | Val_Acc: 63.616\n",
      "Epoch 010 | Train_Loss: 0.64803 | Train_Acc: 62.273 | Val_Loss: 0.621 | Val_Acc: 65.315\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.676735 \n",
      "Train mean acc: 57.389493\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.620713 \n",
      "Validation mean acc: 65.315290\n",
      "\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "cells = ['RNN','LSTM','GRU']\n",
    "hidden_sizes = [64,128,256]\n",
    "\n",
    "for hidden_size in hidden_sizes:\n",
    "  for cell in cells:\n",
    "\n",
    "    print(\"\\n---------  \",cell,\" | \",hidden_size,\"  ------------ \\n\")\n",
    "    net = RNN_TweetClassifier(\n",
    "        vocab_size = len(text_field.vocab),\n",
    "        embedding_dim = glove_length,\n",
    "        input_size = vector_length,\n",
    "        hidden_size = hidden_size,\n",
    "        output_size = 1,\n",
    "        num_layers = 2,\n",
    "        bidirectional = True,\n",
    "        cell_type = 'GRU',\n",
    "        embedding = text_field.vocab.vectors,\n",
    "        dropout_prob = 0.2\n",
    "    )\n",
    "    net\n",
    "    learning_rate = 0.01\n",
    "    epochs = 10\n",
    "    loss_function = nn.BCELoss()\n",
    "    optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "    lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "    args.gradient_clip = 2\n",
    "\n",
    "    train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)\n",
    "    print(\"\\n\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "mfe6Qq0IlE0T"
   },
   "source": [
    "**Remarks**:\n",
    "\n",
    " Hidden size performance depends a lot on the size of the data. For example, for 50.000 tweets hidden size 64 has the best performance. In contrast in the executions that I used more than 200.000 tweets, 128 and 256 hidden sizes had better performance. For this reason I left 128 as the hidden size in the cells above."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "YDc3EMMyWu4G"
   },
   "source": [
    "## __Stacked RNNs__\n",
    "\n",
    "In this section I will experiment with the network structure, which means:\n",
    "\n",
    "* Changing the number of layers,\n",
    "* Adding different stacked RNNs,LSTMs and GRUs\n",
    "* Adding more layers like Linear and ReLU\n",
    "* Changing the last instance and replacing it with element-wise min-max\n",
    "* Skiping connections\n",
    "\n",
    "For all these I will use the best parameters from previous experiments.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "xDZc9F2hjZFt"
   },
   "source": [
    "### Changing the number of layers (stacked RNNs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "-GxMBeMtjdon",
    "outputId": "ccda89e4-71ed-4a00-d7b9-98816070867e"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "---------   LSTM  |  2   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.70127 | Train_Acc: 50.611 | Val_Loss: 0.702 | Val_Acc: 51.057\n",
      "Epoch 002 | Train_Loss: 0.70093 | Train_Acc: 50.167 | Val_Loss: 0.703 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.70029 | Train_Acc: 50.167 | Val_Loss: 0.707 | Val_Acc: 50.823\n",
      "Epoch 004 | Train_Loss: 0.69890 | Train_Acc: 50.205 | Val_Loss: 0.709 | Val_Acc: 49.436\n",
      "Epoch 005 | Train_Loss: 0.69876 | Train_Acc: 50.408 | Val_Loss: 0.697 | Val_Acc: 50.940\n",
      "Epoch 006 | Train_Loss: 0.69710 | Train_Acc: 50.484 | Val_Loss: 0.696 | Val_Acc: 50.798\n",
      "Epoch 007 | Train_Loss: 0.69729 | Train_Acc: 50.174 | Val_Loss: 0.692 | Val_Acc: 51.922\n",
      "Epoch 008 | Train_Loss: 0.69721 | Train_Acc: 50.250 | Val_Loss: 0.695 | Val_Acc: 49.157\n",
      "Epoch 009 | Train_Loss: 0.69573 | Train_Acc: 50.602 | Val_Loss: 0.698 | Val_Acc: 49.060\n",
      "Epoch 010 | Train_Loss: 0.69655 | Train_Acc: 50.658 | Val_Loss: 0.704 | Val_Acc: 51.155\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.698403 \n",
      "Train mean acc: 50.372620\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.704321 \n",
      "Validation mean acc: 51.155134\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  2   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.69903 | Train_Acc: 50.719 | Val_Loss: 0.692 | Val_Acc: 51.911\n",
      "Epoch 002 | Train_Loss: 0.69919 | Train_Acc: 51.563 | Val_Loss: 0.700 | Val_Acc: 50.960\n",
      "Epoch 003 | Train_Loss: 0.69634 | Train_Acc: 51.853 | Val_Loss: 0.691 | Val_Acc: 52.810\n",
      "Epoch 004 | Train_Loss: 0.69818 | Train_Acc: 51.361 | Val_Loss: 0.690 | Val_Acc: 53.532\n",
      "Epoch 005 | Train_Loss: 0.69671 | Train_Acc: 52.034 | Val_Loss: 0.691 | Val_Acc: 53.284\n",
      "Epoch 006 | Train_Loss: 0.69582 | Train_Acc: 52.339 | Val_Loss: 0.702 | Val_Acc: 50.564\n",
      "Epoch 007 | Train_Loss: 0.70005 | Train_Acc: 51.439 | Val_Loss: 0.693 | Val_Acc: 52.762\n",
      "Epoch 008 | Train_Loss: 0.69789 | Train_Acc: 51.645 | Val_Loss: 0.689 | Val_Acc: 53.175\n",
      "Epoch 009 | Train_Loss: 0.69884 | Train_Acc: 51.927 | Val_Loss: 0.691 | Val_Acc: 53.513\n",
      "Epoch 010 | Train_Loss: 0.69817 | Train_Acc: 51.717 | Val_Loss: 0.716 | Val_Acc: 50.921\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.698023 \n",
      "Train mean acc: 51.659462\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.715702 \n",
      "Validation mean acc: 50.920759\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   LSTM  |  3   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.65771 | Train_Acc: 58.932 | Val_Loss: 0.613 | Val_Acc: 66.872\n",
      "Epoch 002 | Train_Loss: 0.60222 | Train_Acc: 67.248 | Val_Loss: 0.582 | Val_Acc: 68.767\n",
      "Epoch 003 | Train_Loss: 0.59099 | Train_Acc: 68.126 | Val_Loss: 0.565 | Val_Acc: 71.292\n",
      "Epoch 004 | Train_Loss: 0.58277 | Train_Acc: 69.065 | Val_Loss: 0.553 | Val_Acc: 71.590\n",
      "Epoch 005 | Train_Loss: 0.57652 | Train_Acc: 69.709 | Val_Loss: 0.552 | Val_Acc: 71.663\n",
      "Epoch 006 | Train_Loss: 0.57280 | Train_Acc: 69.718 | Val_Loss: 0.557 | Val_Acc: 70.179\n",
      "Epoch 007 | Train_Loss: 0.57122 | Train_Acc: 69.703 | Val_Loss: 0.567 | Val_Acc: 69.607\n",
      "Epoch 008 | Train_Loss: 0.57051 | Train_Acc: 69.910 | Val_Loss: 0.596 | Val_Acc: 65.862\n",
      "Epoch 009 | Train_Loss: 0.57340 | Train_Acc: 69.750 | Val_Loss: 0.554 | Val_Acc: 71.643\n",
      "Epoch 010 | Train_Loss: 0.57383 | Train_Acc: 69.388 | Val_Loss: 0.557 | Val_Acc: 71.038\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.587195 \n",
      "Train mean acc: 68.154969\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.556640 \n",
      "Validation mean acc: 71.037946\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  3   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.65548 | Train_Acc: 59.936 | Val_Loss: 0.596 | Val_Acc: 65.017\n",
      "Epoch 002 | Train_Loss: 0.59540 | Train_Acc: 67.782 | Val_Loss: 0.588 | Val_Acc: 69.188\n",
      "Epoch 003 | Train_Loss: 0.57848 | Train_Acc: 69.425 | Val_Loss: 0.549 | Val_Acc: 71.350\n",
      "Epoch 004 | Train_Loss: 0.56816 | Train_Acc: 70.265 | Val_Loss: 0.564 | Val_Acc: 71.239\n",
      "Epoch 005 | Train_Loss: 0.57034 | Train_Acc: 70.339 | Val_Loss: 0.546 | Val_Acc: 71.839\n",
      "Epoch 006 | Train_Loss: 0.56590 | Train_Acc: 70.582 | Val_Loss: 0.547 | Val_Acc: 72.547\n",
      "Epoch 007 | Train_Loss: 0.56689 | Train_Acc: 70.645 | Val_Loss: 0.547 | Val_Acc: 72.176\n",
      "Epoch 008 | Train_Loss: 0.56414 | Train_Acc: 70.466 | Val_Loss: 0.575 | Val_Acc: 71.819\n",
      "Epoch 009 | Train_Loss: 0.57349 | Train_Acc: 70.013 | Val_Loss: 0.573 | Val_Acc: 70.628\n",
      "Epoch 010 | Train_Loss: 0.56747 | Train_Acc: 70.485 | Val_Loss: 0.542 | Val_Acc: 71.975\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.580574 \n",
      "Train mean acc: 68.993760\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.542244 \n",
      "Validation mean acc: 71.975446\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   LSTM  |  5   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.70380 | Train_Acc: 50.030 | Val_Loss: 0.693 | Val_Acc: 49.060\n",
      "Epoch 002 | Train_Loss: 0.70013 | Train_Acc: 50.268 | Val_Loss: 0.764 | Val_Acc: 49.060\n",
      "Epoch 003 | Train_Loss: 0.70125 | Train_Acc: 50.020 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 004 | Train_Loss: 0.70676 | Train_Acc: 49.815 | Val_Loss: 0.712 | Val_Acc: 50.940\n",
      "Epoch 005 | Train_Loss: 0.70501 | Train_Acc: 50.022 | Val_Loss: 0.715 | Val_Acc: 49.060\n",
      "Epoch 006 | Train_Loss: 0.68792 | Train_Acc: 53.611 | Val_Loss: 0.719 | Val_Acc: 62.522\n",
      "Epoch 007 | Train_Loss: 0.64765 | Train_Acc: 63.021 | Val_Loss: 0.642 | Val_Acc: 62.444\n",
      "Epoch 008 | Train_Loss: 0.62475 | Train_Acc: 65.346 | Val_Loss: 0.605 | Val_Acc: 66.448\n",
      "Epoch 009 | Train_Loss: 0.62883 | Train_Acc: 64.842 | Val_Loss: 0.599 | Val_Acc: 66.761\n",
      "Epoch 010 | Train_Loss: 0.61640 | Train_Acc: 66.053 | Val_Loss: 0.593 | Val_Acc: 68.343\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.672250 \n",
      "Train mean acc: 56.302965\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.593165 \n",
      "Validation mean acc: 68.342634\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  5   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.65963 | Train_Acc: 60.844 | Val_Loss: 0.611 | Val_Acc: 65.382\n",
      "Epoch 002 | Train_Loss: 0.62098 | Train_Acc: 65.822 | Val_Loss: 0.603 | Val_Acc: 67.787\n",
      "Epoch 003 | Train_Loss: 0.60784 | Train_Acc: 66.815 | Val_Loss: 0.604 | Val_Acc: 67.709\n",
      "Epoch 004 | Train_Loss: 0.60726 | Train_Acc: 67.082 | Val_Loss: 0.634 | Val_Acc: 61.747\n",
      "Epoch 005 | Train_Loss: 0.60523 | Train_Acc: 66.801 | Val_Loss: 0.588 | Val_Acc: 67.048\n",
      "Epoch 006 | Train_Loss: 0.60602 | Train_Acc: 66.937 | Val_Loss: 0.574 | Val_Acc: 69.774\n",
      "Epoch 007 | Train_Loss: 0.60265 | Train_Acc: 67.412 | Val_Loss: 0.592 | Val_Acc: 67.835\n",
      "Epoch 008 | Train_Loss: 0.60531 | Train_Acc: 67.104 | Val_Loss: 0.586 | Val_Acc: 67.567\n",
      "Epoch 009 | Train_Loss: 0.61685 | Train_Acc: 65.952 | Val_Loss: 0.595 | Val_Acc: 68.412\n",
      "Epoch 010 | Train_Loss: 0.61779 | Train_Acc: 66.149 | Val_Loss: 0.606 | Val_Acc: 66.161\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.614954 \n",
      "Train mean acc: 66.091886\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.605684 \n",
      "Validation mean acc: 66.160714\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   LSTM  |  8   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.70853 | Train_Acc: 50.205 | Val_Loss: 0.701 | Val_Acc: 49.060\n",
      "Epoch 002 | Train_Loss: 0.70231 | Train_Acc: 49.925 | Val_Loss: 0.698 | Val_Acc: 49.060\n",
      "Epoch 003 | Train_Loss: 0.70352 | Train_Acc: 50.244 | Val_Loss: 0.698 | Val_Acc: 50.940\n",
      "Epoch 004 | Train_Loss: 0.69985 | Train_Acc: 50.259 | Val_Loss: 0.698 | Val_Acc: 50.940\n",
      "Epoch 005 | Train_Loss: 0.70263 | Train_Acc: 50.345 | Val_Loss: 0.708 | Val_Acc: 50.940\n",
      "Epoch 006 | Train_Loss: 0.70141 | Train_Acc: 49.858 | Val_Loss: 0.701 | Val_Acc: 49.060\n",
      "Epoch 007 | Train_Loss: 0.70352 | Train_Acc: 49.869 | Val_Loss: 0.718 | Val_Acc: 50.940\n",
      "Epoch 008 | Train_Loss: 0.70194 | Train_Acc: 50.027 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 009 | Train_Loss: 0.69900 | Train_Acc: 50.158 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 010 | Train_Loss: 0.70209 | Train_Acc: 49.648 | Val_Loss: 0.694 | Val_Acc: 49.060\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.702480 \n",
      "Train mean acc: 50.053847\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.693875 \n",
      "Validation mean acc: 49.059710\n",
      "\n",
      "\n",
      "\n",
      "\n",
      "---------   GRU  |  8   ------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.70825 | Train_Acc: 49.985 | Val_Loss: 0.699 | Val_Acc: 50.940\n",
      "Epoch 002 | Train_Loss: 0.70513 | Train_Acc: 49.835 | Val_Loss: 0.695 | Val_Acc: 50.940\n",
      "Epoch 003 | Train_Loss: 0.69914 | Train_Acc: 49.863 | Val_Loss: 0.696 | Val_Acc: 50.940\n",
      "Epoch 004 | Train_Loss: 0.70046 | Train_Acc: 49.897 | Val_Loss: 0.726 | Val_Acc: 49.060\n",
      "Epoch 005 | Train_Loss: 0.70401 | Train_Acc: 50.125 | Val_Loss: 0.695 | Val_Acc: 50.940\n",
      "Epoch 006 | Train_Loss: 0.69849 | Train_Acc: 50.126 | Val_Loss: 0.693 | Val_Acc: 49.060\n",
      "Epoch 007 | Train_Loss: 0.70019 | Train_Acc: 50.162 | Val_Loss: 0.702 | Val_Acc: 49.060\n",
      "Epoch 008 | Train_Loss: 0.70265 | Train_Acc: 50.008 | Val_Loss: 0.693 | Val_Acc: 50.940\n",
      "Epoch 009 | Train_Loss: 0.70436 | Train_Acc: 49.865 | Val_Loss: 0.693 | Val_Acc: 49.060\n",
      "Epoch 010 | Train_Loss: 0.69946 | Train_Acc: 50.215 | Val_Loss: 0.697 | Val_Acc: 50.940\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.702214 \n",
      "Train mean acc: 50.008170\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.696920 \n",
      "Validation mean acc: 50.940290\n",
      "\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "cells = ['LSTM','GRU']\n",
    "layers = [2,3,5,8]\n",
    "\n",
    "for layer in layers:\n",
    "  for cell in cells:\n",
    "\n",
    "    print(\"\\n---------  \",cell,\" | \",layer,\"  ------------ \\n\")\n",
    "    net = RNN_TweetClassifier(\n",
    "        vocab_size = len(text_field.vocab),\n",
    "        embedding_dim = glove_length,\n",
    "        input_size = vector_length,\n",
    "        hidden_size = 128,\n",
    "        output_size = 1,\n",
    "        num_layers = layer,\n",
    "        bidirectional = True,\n",
    "        cell_type = 'GRU',\n",
    "        embedding = text_field.vocab.vectors,\n",
    "        dropout_prob = 0.2\n",
    "    )\n",
    "    net\n",
    "    learning_rate = 0.01\n",
    "    epochs = 10\n",
    "    loss_function = nn.BCELoss()\n",
    "    optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "    lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "    args.gradient_clip = 2\n",
    "\n",
    "    train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)\n",
    "    print(\"\\n\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "lHaNo8wln2mt"
   },
   "source": [
    "**Remarks:** \n",
    "\n",
    "I know that more dense NNs, better perform with big data. For 50.000 tweets I can clearly sεe that the model need less layers and for less layers performs better. With num layers bigger than 2 the scores decrease and the performance is not good. For 1.000.000 tweets that I ve tried , num layers increase, had a serious impact on performance. \n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "xrzhAbzfE26X"
   },
   "source": [
    "## Last, Average, Element-wise min and max of hidden states\n",
    "\n",
    "In this section I am going to experiment with the output hidden state selection"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "aXQQMEK-MZum"
   },
   "source": [
    "Re-instantiating the network:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "oBXemAQfW5bh"
   },
   "outputs": [],
   "source": [
    "class RNN_TweetClassifier(nn.Module):\n",
    "    def __init__(self, cell_type, input_size, output_size, vocab_size, embedding_dim, hidden_size,num_layers, num_classes=1,hidden_dim=200,hidden_size1 = 500,hidden_size2 = 100, embedding=None ,bidirectional=True ,dropout_prob=0.2,output_state='Last'):\n",
    "        super(RNN_TweetClassifier,self).__init__()\n",
    "\n",
    "\n",
    "        cells = {\n",
    "          \"RNN\"     : nn.RNN,\n",
    "          \"LSTM\"    : nn.LSTM,\n",
    "          \"GRU\"     : nn.GRU\n",
    "        }\n",
    "\n",
    "        self.cell_type = cell_type\n",
    "        self.dropout_prob = dropout_prob\n",
    "        self.output_state = output_state\n",
    "        self.input_size, self.output_size, self.hidden_size, self.vocab_size, self.embedding_dim ,self.num_layers = input_size, output_size, hidden_size, vocab_size, embedding_dim, num_layers\n",
    "        self.bidirectional = bidirectional\n",
    "\n",
    "        self.embedding = nn.Embedding.from_pretrained(torch.FloatTensor(embedding),padding_idx=0)\n",
    "        \n",
    "        self.rnn = cells[cell_type](         # Pick the specific model\n",
    "            input_size=embedding_dim,           # Number of features for each time step\n",
    "            hidden_size=hidden_size,         # Rnn hidden embedding representation\n",
    "            num_layers=num_layers,           # Number of layers (if >1 it is a stacked RNN)\n",
    "            bidirectional=bidirectional,     # If true it is a \n",
    "            batch_first=True,                # input & output will have batch size as 1st dimension. e.g. (batch, time_step, input_size)\n",
    "            dropout = dropout_prob if dropout_prob != None else 0,\n",
    "\n",
    "        )        \n",
    "\n",
    "        if dropout_prob != None:\n",
    "          self.dropout = nn.Dropout(p = dropout_prob)\n",
    "\n",
    "        if bidirectional == True:\n",
    "          self.fc1 = nn.Linear(hidden_size*2 , output_size)        \n",
    "        else:\n",
    "          self.fc1 = nn.Linear(hidden_size , output_size)\n",
    "\n",
    "\n",
    "        self.sigmoid = nn.Sigmoid()\n",
    "        \n",
    "    def forward(self, x):\n",
    "        \n",
    "        out = self.embedding(x)\n",
    "        \n",
    "        if self.dropout_prob != None:\n",
    "          out = self.dropout(out)\n",
    "\n",
    "        if self.cell_type == 'LSTM':\n",
    "            r_out, (last_hidden_state, last_cell_state) = self.rnn(out.detach()) \n",
    "        else:\n",
    "            r_out, last_hidden_state = self.rnn(out.detach())\n",
    "\n",
    "      \n",
    "        # ------   STATE  ------\n",
    "        if self.output_state == 'Last' :\n",
    "          r_out = r_out[:, -1, :]\n",
    "\n",
    "        elif self.output_state == 'Mean': \n",
    "          r_out = torch.mean(r_out,1)\n",
    "        \n",
    "        elif self.output_state == 'Max':\n",
    "          r_out,_ = torch.max(r_out,dim=1)\n",
    "        \n",
    "        elif self.output_state == 'Min':\n",
    "          r_out,_ = torch.min(r_out,dim=1)\n",
    "        \n",
    "      \n",
    "        out = self.fc1(r_out) \n",
    "        out = self.sigmoid(out)\n",
    "\n",
    "        return out"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "kWWbjyhasgFn"
   },
   "source": [
    "I will experiment with GRU, with the same parameters but with different last state. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "7aTLAgjOlINL",
    "outputId": "35748f37-536f-4d9c-f24c-c90d309732ce"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "------------------   Last   ------------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.65243 | Train_Acc: 60.877 | Val_Loss: 0.601 | Val_Acc: 67.171\n",
      "Epoch 002 | Train_Loss: 0.63158 | Train_Acc: 64.538 | Val_Loss: 0.577 | Val_Acc: 70.003\n",
      "Epoch 003 | Train_Loss: 0.59549 | Train_Acc: 68.081 | Val_Loss: 0.604 | Val_Acc: 67.620\n",
      "Epoch 004 | Train_Loss: 0.59074 | Train_Acc: 68.767 | Val_Loss: 0.576 | Val_Acc: 68.041\n",
      "Epoch 005 | Train_Loss: 0.57681 | Train_Acc: 69.740 | Val_Loss: 0.576 | Val_Acc: 69.247\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.609410 \n",
      "Train mean acc: 66.400359\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.576219 \n",
      "Validation mean acc: 69.246652\n",
      "\n",
      "------------------   Min   ------------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.66726 | Train_Acc: 60.491 | Val_Loss: 0.606 | Val_Acc: 66.233\n",
      "Epoch 002 | Train_Loss: 0.61456 | Train_Acc: 66.514 | Val_Loss: 0.576 | Val_Acc: 69.456\n",
      "Epoch 003 | Train_Loss: 0.59143 | Train_Acc: 68.318 | Val_Loss: 0.619 | Val_Acc: 65.778\n",
      "Epoch 004 | Train_Loss: 0.57430 | Train_Acc: 70.003 | Val_Loss: 0.540 | Val_Acc: 73.016\n",
      "Epoch 005 | Train_Loss: 0.57891 | Train_Acc: 69.922 | Val_Loss: 0.529 | Val_Acc: 73.133\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.605292 \n",
      "Train mean acc: 67.049587\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.528949 \n",
      "Validation mean acc: 73.133371\n",
      "\n",
      "------------------   Max   ------------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.60836 | Train_Acc: 66.733 | Val_Loss: 0.583 | Val_Acc: 69.079\n",
      "Epoch 002 | Train_Loss: 0.56838 | Train_Acc: 70.262 | Val_Loss: 0.540 | Val_Acc: 71.961\n",
      "Epoch 003 | Train_Loss: 0.56454 | Train_Acc: 70.319 | Val_Loss: 0.546 | Val_Acc: 71.423\n",
      "Epoch 004 | Train_Loss: 0.55960 | Train_Acc: 70.833 | Val_Loss: 0.549 | Val_Acc: 71.331\n",
      "Epoch 005 | Train_Loss: 0.57348 | Train_Acc: 69.813 | Val_Loss: 0.542 | Val_Acc: 71.624\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.574872 \n",
      "Train mean acc: 69.592119\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.541779 \n",
      "Validation mean acc: 71.623884\n",
      "\n",
      "------------------   Mean   ------------------ \n",
      "\n",
      "Epoch 001 | Train_Loss: 0.66105 | Train_Acc: 60.852 | Val_Loss: 0.622 | Val_Acc: 65.335\n",
      "Epoch 002 | Train_Loss: 0.62798 | Train_Acc: 64.731 | Val_Loss: 0.654 | Val_Acc: 61.272\n",
      "Epoch 003 | Train_Loss: 0.61830 | Train_Acc: 65.932 | Val_Loss: 0.593 | Val_Acc: 67.991\n",
      "Epoch 004 | Train_Loss: 0.60144 | Train_Acc: 67.410 | Val_Loss: 0.601 | Val_Acc: 66.409\n",
      "Epoch 005 | Train_Loss: 0.60358 | Train_Acc: 67.575 | Val_Loss: 0.552 | Val_Acc: 72.118\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.622467 \n",
      "Train mean acc: 65.300020\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.551579 \n",
      "Validation mean acc: 72.117746\n"
     ]
    }
   ],
   "source": [
    "out_states = ['Last','Min','Max','Mean']\n",
    "\n",
    "for state in out_states:\n",
    "  print(\"\\n------------------  \",state,\"  ------------------ \\n\")\n",
    "  net = RNN_TweetClassifier(\n",
    "      vocab_size = len(text_field.vocab),\n",
    "      embedding_dim = glove_length,\n",
    "      input_size = vector_length,\n",
    "      hidden_size = 256,\n",
    "      output_size = 1,\n",
    "      num_layers = 2,\n",
    "      bidirectional = True,\n",
    "      cell_type = 'GRU',\n",
    "      embedding = text_field.vocab.vectors,\n",
    "      dropout_prob = 0.2,\n",
    "      output_state = state\n",
    "  )\n",
    "  net\n",
    "  learning_rate = 0.01\n",
    "  epochs = 5\n",
    "  loss_function = nn.BCELoss()\n",
    "  optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "  lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "  args.gradient_clip = 5\n",
    "\n",
    "  train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "KzFAhis4tOsS"
   },
   "source": [
    "**Remarks:** I can see that element-wise mean has a good performance. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "X5j2hsLUGF6k"
   },
   "source": [
    "## __Skip connections__\n",
    "\n",
    "A \"shortcut\" or a \"skip connection\" allows the gradient to be directly backpropagated to earlier layers.\n",
    "\n",
    "Building a danse neural net based on ResNet and GoogLeNet. So I am going to use a BatchNorm1d layer.\n",
    "\n",
    "I tried to implement a net like the ResNet shown in the picture. \n",
    " [Source Link](https://d2l.ai/chapter_convolutional-modern/resnet.html?highlight=skip%20connections)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "tMfCnL3nl45A"
   },
   "source": [
    "<img src=\"https://d2l.ai/_images/resnet-block.svg\"/>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "vdIMHhFzXESj"
   },
   "outputs": [],
   "source": [
    "class RNN_TweetClassifier(nn.Module):\n",
    "    def __init__(self, cell_type, input_size, output_size, vocab_size, embedding_dim, hidden_size,num_layers, num_classes=1,hidden_dim=200,hidden_size1 = 500,hidden_size2 = 100, embedding=None ,bidirectional=True ,dropout_prob=0.2,output_state='Last'):\n",
    "        super(RNN_TweetClassifier,self).__init__()\n",
    "\n",
    "\n",
    "        cells = {\n",
    "          \"RNN\"     : nn.RNN,\n",
    "          \"LSTM\"    : nn.LSTM,\n",
    "          \"GRU\"     : nn.GRU\n",
    "        }\n",
    "\n",
    "        self.cell_type = cell_type\n",
    "        self.dropout_prob = dropout_prob\n",
    "        self.output_state = output_state\n",
    "        self.input_size, self.output_size, self.hidden_size, self.vocab_size, self.embedding_dim ,self.num_layers = input_size, output_size, hidden_size, vocab_size, embedding_dim, num_layers\n",
    "        self.bidirectional = bidirectional\n",
    "\n",
    "        self.embedding = nn.Embedding.from_pretrained(torch.FloatTensor(embedding),padding_idx=0)\n",
    "\n",
    "\n",
    "        self.block1 = nn.Sequential(\n",
    "            cells[cell_type](           # Pick the specific model\n",
    "                        input_size=2*hidden_size,                  # Number of features for each time step\n",
    "                        hidden_size=hidden_size,                 # Rnn hidden embedding representation\n",
    "                        num_layers=2,                            # Number of layers (if >1 it is a stacked RNN)\n",
    "                        bidirectional=bidirectional,             # If true it is a \n",
    "                        batch_first=True,                        # input & output will have batch size as 1st dimension. e.g. (batch, time_step, input_size)\n",
    "                        dropout = dropout_prob if dropout_prob != None else 0,\n",
    "              ),\n",
    "              nn.BatchNorm1d(input_size, affine=False),\n",
    "              nn.Dropout(p = dropout_prob),\n",
    "              nn.ReLU()\n",
    "        )\n",
    "\n",
    "        self.normalizeLayer = nn.BatchNorm1d((input_size), affine=False)\n",
    "\n",
    "        self.relu =  nn.ReLU()\n",
    "\n",
    "        # THE INITIAL HIDDEN STATE \n",
    "        self.rnn = cells[cell_type](          # Pick the specific model\n",
    "            input_size=embedding_dim,         # Number of features for each time step\n",
    "            hidden_size=hidden_size,          # Rnn hidden embedding representation\n",
    "            num_layers=2,                     # Number of layers (if >1 it is a stacked RNN)\n",
    "            bidirectional=bidirectional,      # If true it is a \n",
    "            batch_first=True,                 # input & output will have batch size as 1st dimension. e.g. (batch, time_step, input_size)\n",
    "            dropout = dropout_prob if dropout_prob != None else 0,\n",
    "        )        \n",
    "\n",
    "        # THE EMBEDDED HIDDEN STATE \n",
    "        self.rnnHidden = cells[cell_type](           # Pick the specific model\n",
    "            input_size=2*hidden_size,                  # Number of features for each time step\n",
    "            hidden_size=hidden_size,                 # Rnn hidden embedding representation\n",
    "            num_layers=2,                            # Number of layers (if >1 it is a stacked RNN)\n",
    "            bidirectional=bidirectional,             # If true it is a \n",
    "            batch_first=True,                        # input & output will have batch size as 1st dimension. e.g. (batch, time_step, input_size)\n",
    "            dropout = dropout_prob if dropout_prob != None else 0,\n",
    "        )      \n",
    "\n",
    "        if dropout_prob != None:\n",
    "          self.dropout = nn.Dropout(p = dropout_prob)\n",
    "\n",
    "        if bidirectional == True:\n",
    "          self.fc1 = nn.Linear(hidden_size*2 , output_size)        \n",
    "        else:\n",
    "          self.fc1 = nn.Linear(hidden_size , output_size)\n",
    "\n",
    "        self.sigmoid = nn.Sigmoid()\n",
    "        \n",
    "    def forward(self, x):\n",
    "\n",
    "        out = self.embedding(x)\n",
    "        \n",
    "        if self.dropout_prob != None:\n",
    "          out = self.dropout(out)\n",
    "\n",
    "        rnns = nn.ModuleList()\n",
    "        for i in range(self.num_layers):\n",
    "          if i == 0:\n",
    "            if self.cell_type == 'LSTM':\n",
    "                r_out, (last_hidden_state, last_cell_state) = self.rnn(out.detach()) \n",
    "                # rnns.append((r_out, (last_hidden_state, last_cell_state)))\n",
    "            else:\n",
    "                r_out, last_hidden_state = self.rnn(out.detach())                \n",
    "                # rnns.append((r_out, last_hidden_state)) \n",
    "            # self.normalizeLayer(r_out)\n",
    "            # r_out = self.relu(r_out)\n",
    "            out = r_out\n",
    "\n",
    "          else :\n",
    "            if self.cell_type == 'LSTM':\n",
    "                r_out, (last_hidden_state, last_cell_state) = self.rnnHidden(r_out.detach()) \n",
    "                # rnns.append((r_out, (last_hidden_state, last_cell_state)))\n",
    "            else:\n",
    "                r_out, last_hidden_state = self.rnnHidden(r_out.detach())\n",
    "                # rnns.append((r_out, last_hidden_state))     \n",
    "\n",
    "            # self.normalizeLayer(r_out)\n",
    "            # r_out = self.relu(r_out)            \n",
    "\n",
    "            # EITHER WITH ADDING OUTPUTS\n",
    "            # r_out += out\n",
    "\n",
    "            # EITHER WITH CONCATING OUTPUTS\n",
    "            r_out = torch.cat((r_out,out.detach()),dim=1)\n",
    "\n",
    "\n",
    "\n",
    "        if self.output_state == 'Last' :\n",
    "          r_out = r_out[:, -1, :]\n",
    "\n",
    "        elif self.output_state == 'Mean': \n",
    "          r_out = torch.mean(r_out,1)\n",
    "        \n",
    "        elif self.output_state == 'Max':\n",
    "          r_out,_ = torch.max(r_out,dim=1)\n",
    "        \n",
    "        elif self.output_state == 'Min':\n",
    "          r_out,_ = torch.min(r_out,dim=1)\n",
    "        \n",
    "      \n",
    "\n",
    "        out = self.fc1(r_out) \n",
    "\n",
    "        # out = self.fc2(out)\n",
    "\n",
    "        out = self.sigmoid(out)\n",
    "\n",
    "        return out"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "_q_0miz8uual"
   },
   "source": [
    "Checking the model with skip connections model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "k9mfGZI9Ac7Z",
    "outputId": "7b7eb716-16a2-4578-a4cc-0c48bf89e5ce"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 001 | Train_Loss: 0.69657 | Train_Acc: 51.258 | Val_Loss: 0.684 | Val_Acc: 54.495\n",
      "Epoch 002 | Train_Loss: 0.64735 | Train_Acc: 62.361 | Val_Loss: 0.606 | Val_Acc: 67.645\n",
      "Epoch 003 | Train_Loss: 0.62709 | Train_Acc: 64.499 | Val_Loss: 0.614 | Val_Acc: 66.677\n",
      "Epoch 004 | Train_Loss: 0.62101 | Train_Acc: 65.226 | Val_Loss: 0.608 | Val_Acc: 66.708\n",
      "Epoch 005 | Train_Loss: 0.62058 | Train_Acc: 65.142 | Val_Loss: 0.607 | Val_Acc: 67.868\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.642521 \n",
      "Train mean acc: 61.697484\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.607163 \n",
      "Validation mean acc: 67.868304\n"
     ]
    }
   ],
   "source": [
    "net = RNN_TweetClassifier(\n",
    "    vocab_size = len(text_field.vocab),\n",
    "    embedding_dim = glove_length,\n",
    "    input_size = vector_length,\n",
    "    hidden_size = 128,\n",
    "    output_size = 1,\n",
    "    num_layers = 5,\n",
    "    bidirectional = True,\n",
    "    cell_type = 'GRU',\n",
    "    embedding = text_field.vocab.vectors,\n",
    "    dropout_prob = 0.2,\n",
    "    output_state = 'Mean'\n",
    ")\n",
    "net\n",
    "learning_rate = 0.01\n",
    "epochs = 5\n",
    "loss_function = nn.BCELoss()\n",
    "optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "args.gradient_clip = 5\n",
    "\n",
    "train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Qi5BIQwHu5hU"
   },
   "source": [
    "**Remarks:** Skip connections are used to dense nets mostly, which means big data. For 50.000 tweets the net can't be much dense. So, the performance wasn't very good."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "FZ0FfOGvvtUV"
   },
   "source": [
    "Re-instantiating the network with more layers as in the Resnet model. I added:\n",
    "\n",
    "- Normalization layer\n",
    "- ReLU\n",
    "- Dropout layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "0cg2l5U2AUn6"
   },
   "outputs": [],
   "source": [
    "class RNN_TweetClassifier(nn.Module):\n",
    "    def __init__(self, cell_type, input_size, output_size, vocab_size, embedding_dim, hidden_size,num_layers, num_classes=1,hidden_dim=200,hidden_size1 = 500,hidden_size2 = 100, embedding=None ,bidirectional=True ,dropout_prob=0.2,output_state='Last'):\n",
    "        super(RNN_TweetClassifier,self).__init__()\n",
    "\n",
    "\n",
    "        cells = {\n",
    "          \"RNN\"     : nn.RNN,\n",
    "          \"LSTM\"    : nn.LSTM,\n",
    "          \"GRU\"     : nn.GRU\n",
    "        }\n",
    "\n",
    "        self.cell_type = cell_type\n",
    "        self.dropout_prob = dropout_prob\n",
    "        self.output_state = output_state\n",
    "        self.input_size, self.output_size, self.hidden_size, self.vocab_size, self.embedding_dim ,self.num_layers = input_size, output_size, hidden_size, vocab_size, embedding_dim, num_layers\n",
    "        self.bidirectional = bidirectional\n",
    "\n",
    "        self.embedding = nn.Embedding.from_pretrained(torch.FloatTensor(embedding),padding_idx=0)\n",
    "\n",
    "\n",
    "        self.block1 = nn.Sequential(\n",
    "            cells[cell_type](                                    # Pick the specific model\n",
    "                        input_size=2*hidden_size,                # Number of features for each time step\n",
    "                        hidden_size=hidden_size,                 # Rnn hidden embedding representation\n",
    "                        num_layers=2,                            # Number of layers (if >1 it is a stacked RNN)\n",
    "                        bidirectional=bidirectional,             # If true it is a \n",
    "                        batch_first=True,                        # input & output will have batch size as 1st dimension. e.g. (batch, time_step, input_size)\n",
    "                        dropout = dropout_prob if dropout_prob != None else 0,\n",
    "              ),\n",
    "              nn.BatchNorm1d(input_size, affine=False),\n",
    "              nn.ReLU(),\n",
    "              nn.Dropout(p = dropout_prob),\n",
    "        )\n",
    "\n",
    "        self.normalizeLayer = nn.BatchNorm1d(input_size, affine=False)\n",
    "\n",
    "        self.relu =  nn.ReLU()\n",
    "\n",
    "        self.rnn = cells[cell_type](          # Pick the specific model\n",
    "            input_size=embedding_dim,         # Number of features for each time step\n",
    "            hidden_size=hidden_size,          # Rnn hidden embedding representation\n",
    "            num_layers=1,                     # Number of layers (if >1 it is a stacked RNN)\n",
    "            bidirectional=bidirectional,      # If true it is a \n",
    "            batch_first=True,                 # input & output will have batch size as 1st dimension. e.g. (batch, time_step, input_size)\n",
    "            dropout =  0,\n",
    "        )        \n",
    "\n",
    "        self.rnnHidden = cells[cell_type](           # Pick the specific model\n",
    "            input_size=2*hidden_size,                  # Number of features for each time step\n",
    "            hidden_size=hidden_size,                 # Rnn hidden embedding representation\n",
    "            num_layers=1,                            # Number of layers (if >1 it is a stacked RNN)\n",
    "            bidirectional=bidirectional,             # If true it is a \n",
    "            batch_first=True,                        # input & output will have batch size as 1st dimension. e.g. (batch, time_step, input_size)\n",
    "            dropout = 0,\n",
    "        )      \n",
    "\n",
    "        if dropout_prob != None:\n",
    "          self.dropout = nn.Dropout(p = dropout_prob)\n",
    "\n",
    "        if bidirectional == True:\n",
    "          self.fc1 = nn.Linear(hidden_size*2 , output_size)        \n",
    "        else:\n",
    "          self.fc1 = nn.Linear(hidden_size , output_size)\n",
    "\n",
    "\n",
    "        # self.fc1 = nn.Linear(input_size* embedding_dim,hidden_size1)\n",
    "        # self.relu1 = nn.ReLU()\n",
    "        \n",
    "        # self.fc2 = nn.Linear(50,1)\n",
    "\n",
    "        self.sigmoid = nn.Sigmoid()\n",
    "        \n",
    "    def forward(self, x):\n",
    "        \n",
    "        out = self.embedding(x)\n",
    "        \n",
    "        if self.dropout_prob != None:\n",
    "          out = self.dropout(out)\n",
    "\n",
    "\n",
    "        for i in range(self.num_layers):\n",
    "\n",
    "          # if first loop\n",
    "          if i == 0:\n",
    "            if self.cell_type == 'LSTM':\n",
    "                r_out, (last_hidden_state, last_cell_state) = self.rnn(out.detach()) \n",
    "            else:\n",
    "                r_out, last_hidden_state = self.rnn(out.detach())\n",
    "            r_out = self.normalizeLayer(r_out)\n",
    "            r_out = self.relu(r_out)\n",
    "            r_out = self.dropout(r_out)\n",
    "            out = r_out\n",
    "\n",
    "          else :\n",
    "            if self.cell_type == 'LSTM':\n",
    "                r_out, (last_hidden_state, last_cell_state) = self.rnnHidden(r_out.detach()) \n",
    "            else:\n",
    "                r_out, last_hidden_state = self.rnnHidden(r_out.detach())\n",
    "            \n",
    "            r_out = self.normalizeLayer(r_out)\n",
    "            r_out = self.relu(r_out)\n",
    "            r_out = self.dropout(r_out)\n",
    "            \n",
    "            # EITHER WITH ADDING OUTPUTS\n",
    "            r_out += out\n",
    "\n",
    "            # EITHER WITH CONCATING OUTPUTS\n",
    "            # r_out = torch.cat((r_out,out),dim=1)\n",
    "\n",
    "\n",
    "        if self.output_state == 'Last' :\n",
    "          r_out = r_out[:, -1, :]\n",
    "\n",
    "        elif self.output_state == 'Mean': \n",
    "          r_out = torch.mean(r_out,1)\n",
    "        \n",
    "        elif self.output_state == 'Max':\n",
    "          r_out,_ = torch.max(r_out,dim=1)\n",
    "        \n",
    "        elif self.output_state == 'Min':\n",
    "          r_out,_ = torch.min(r_out,dim=1)\n",
    "\n",
    "        out = self.fc1(r_out) \n",
    "        out = self.sigmoid(out)\n",
    "\n",
    "        return out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "1tp21jypoEB_",
    "outputId": "85c642f7-c0c3-4a26-ea45-288033e7f2c3"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 001 | Train_Loss: 0.72991 | Train_Acc: 52.818 | Val_Loss: 0.684 | Val_Acc: 51.702\n",
      "Epoch 002 | Train_Loss: 0.66485 | Train_Acc: 60.063 | Val_Loss: 0.771 | Val_Acc: 49.118\n",
      "Epoch 003 | Train_Loss: 0.66347 | Train_Acc: 60.406 | Val_Loss: 0.664 | Val_Acc: 56.155\n",
      "Epoch 004 | Train_Loss: 0.64535 | Train_Acc: 62.912 | Val_Loss: 0.627 | Val_Acc: 66.956\n",
      "Epoch 005 | Train_Loss: 0.64987 | Train_Acc: 62.660 | Val_Loss: 0.618 | Val_Acc: 67.112\n",
      "\n",
      "TRAIN STATS:\n",
      "Train mean loss: 0.670690 \n",
      "Train mean acc: 59.771699\n",
      "\n",
      "VALIDATION STATS:\n",
      "Validation mean loss: 0.618174 \n",
      "Validation mean acc: 67.112165\n"
     ]
    }
   ],
   "source": [
    "net = RNN_TweetClassifier(\n",
    "    vocab_size    = len(text_field.vocab),\n",
    "    embedding_dim = glove_length,\n",
    "    input_size    = vector_length,\n",
    "    hidden_size   = 256,\n",
    "    output_size   = 1,\n",
    "    num_layers    = 10,\n",
    "    bidirectional = True,\n",
    "    cell_type     = 'GRU',\n",
    "    embedding     = text_field.vocab.vectors,\n",
    "    dropout_prob  = 0.2,\n",
    "    output_state  = 'Mean'\n",
    ")\n",
    "learning_rate = 0.01\n",
    "epochs = 5\n",
    "loss_function = nn.BCELoss()\n",
    "optimizer =  Adam(net.parameters(),lr = learning_rate)\n",
    "lr_scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer)\n",
    "\n",
    "args.gradient_clip = 5\n",
    "\n",
    "train_loss,train_acc,val_loss,val_acc = train_and_validate(net,epochs,loss_function,optimizer,lr_scheduler,args)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "37nu5shbwPMD"
   },
   "source": [
    "**Remarks:** Skip connections model like ResNet didn't have so good results, as the model got to be much more dense. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "MHNYo6MujU8A"
   },
   "source": [
    "# Overall __Remarks__\n",
    "---\n",
    "\n",
    "###__Type of cell__ \n",
    "\n",
    "RNNs had a good performance, but in almost every execution, the time executing was big. LSTN and GRUs were quicker but ecpecially LSTM and RNN had not the same good performance in the majority of the executions. I noticed that LSTM and RNN didn't even learn in many executions in comparison to the GRU cells.\n",
    "I think that GRU had in general the biggest scores, a \"smooth\" learning rate and a descent time of execution.\n",
    "\n",
    "I noticed in multiple executions overfiting.  \n",
    "\n",
    "###__Gradient clipping__\n",
    "> Gradient clipping is a technique that tackles exploding gradients. The idea of gradient clipping is very simple: If the gradient gets too large, we rescale it to keep it small.\n",
    "\n",
    "Gradient clipping is vital to my executions as in the first three executions results were not so good (partly because of the exploding gradients) and at the time I added the graddient clipping, resuts and learning rate were improved a lot. (except from GRU which had some good results even without gradient clipping)\n",
    "\n",
    "###__Dropout probability__\n",
    "\n",
    "As part of tackling overfitting, dropout layer used in the most NNs, had also some good impact as it make the score increase and loss decrease much smoother.\n",
    "This happens only for dropout sizes like 0.2,0.1 and not for bigger values, as I expected it.\n",
    "\n",
    "\n",
    "###__Hidden size__\n",
    "\n",
    "Bigger hidden sizes,  made the nets much denser and as a result I noticed overfitting. For my executions with 50.000 tweets (that it will be presented in this notebook), it seemed that data were less that those needed from a dense NN. In others executions that I increased the data size and the vector lengths, more dense NNs were needed, or more epochs in the not so dense NNs. Overall, hidden size was a factor that made the results better or worse in dependant with the data size. \n",
    "\n",
    "###__Stacked RNNs__\n",
    "> It is common for larger networks (more layers or more nodes) to more easily overfit the training data.\n",
    "\n",
    "Same as before, increase at layers made the network more dense and hence it mostly overffited the data for num_layers bigger that 3.\n",
    "\n",
    "\n",
    "###__Last output state__\n",
    "\n",
    "Element-wise mean and also picking the last state had some good results .\n",
    "\n",
    "\n",
    "###__Skip connections__\n",
    "\n",
    "This model, took me the most of the time as I didn't know which cobination of skip connections I should use. In most of my tries I made the net much denser and I had the same results as mentioned before. At the end I tried to mimicate the ResNet and GoogleNet in a link in MNIST colab. These gave me some promising results but still not as good as I had in the more simple NNs in the previous experiments I made. \n",
    "\n",
    "---\n",
    "# Best model in SentimentClassiffier_GloVe_RNN_BestModel.ipynb\n",
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "V_vvQqhZqaat"
   },
   "source": [
    "# References\n",
    "\n",
    "[1] Deep Learning For NLP with PyTorch and Torchtext [TowardsDataScience](https://towardsdatascience.com/deep-learning-for-nlp-with-pytorch-and-torchtext-4f92d69052f) \n",
    "\n",
    "[2] Deep Learning For NLP with PyTorch and Torchtext  [dl-for-nlp-pytorch-torchtext.ipynb](https://colab.research.google.com/github/agungsantoso/deep-learning-v2-pytorch/blob/master/sentiment-rnn/Sentiment_RNN_Exercise.ipynb#scrollTo=AVzirwGqpmva)\n",
    "\n",
    "[3] \n",
    "Sentiment Analysis with an RNN [Sentiment_RNN_Exercise.ipynb](https://colab.research.google.com/github/agungsantoso/deep-learning-v2-pytorch/blob/master/sentiment-rnn/Sentiment_RNN_Exercise.ipynb#scrollTo=AVzirwGqpmva)\n",
    "\n",
    "[4] NLP-Bginner-RNN-GRU-LSTM-Embeddings-GloVe-using-PyTorch [GitHub](https://github.com/deshwalmahesh/NLP-Bginner-RNN-GRU-LSTM-Embeddings-GloVe-using-PyTorch/blob/master/nlp-beginner-1-rnn-lstm-gru-embeddings-glove.ipynb)\n",
    "\n",
    "[5] imdb-sentiment-analysis-with-pytorch [GitHub](https://github.com/ShalyginaA/imdb-sentiment-analysis-with-pytorch/blob/master/imdb_analysis.ipynb)\n",
    "\n",
    "[6]How to use Pre-trained Word Embeddings in PyTorch [Medium](https://medium.com/@martinpella/how-to-use-pre-trained-word-embeddings-in-pytorch-71ca59249f76)\n",
    "\n",
    "[7] PyTorch: Loading word vectors into Field vocabulary vs. Embedding layer [StackOverflow](https://stackoverflow.com/questions/62291303/pytorch-loading-word-vectors-into-field-vocabulary-vs-embedding-layer)"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [
    "vc56d1wI5lMz",
    "9wp8npXJgBSh",
    "4g66u1IinG_w"
   ],
   "name": "SentimentClassiffier_GloVe_RNN",
   "provenance": [],
   "toc_visible": true
  },
  "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}