data_augmentation_investigation.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/timmonkey/Desktop/Imperial/Spring Term/Natural Language Processing/Coursework/nlp_classification/venv/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "from datasets import Dataset\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load data\n",
    "train = Dataset.load_from_disk(\"data/train\")\n",
    "val = Dataset.load_from_disk(\"data/val\")\n",
    "dev = Dataset.load_from_disk(\"data/dev\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# only want the text and label columns\n",
    "train_val_df = pd.concat([pd.DataFrame(train), pd.DataFrame(val)])[['text', 'pcl']]\n",
    "test_df = pd.DataFrame(dev)[['text', 'pcl']]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Apply back-translation Augmenter"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.feature_extraction.text import CountVectorizer\n",
    "\n",
    "pos_tokens = CountVectorizer().fit_transform(train_val_df[train_val_df.pcl == 1].text.values)\n",
    "pos_num_tokens = np.sum(pos_tokens.todense(), axis=1).flat\n",
    "\n",
    "\n",
    "neg_tokens = CountVectorizer().fit_transform(train_val_df[train_val_df.pcl == 0].text.values)\n",
    "neg_num_tokens = np.sum(neg_tokens.todense(), axis=1).flat\n",
    "\n",
    "plt.hist(pos_num_tokens, density=True, alpha=0.5, bins='auto', label='pos_tokens')\n",
    "plt.hist(neg_num_tokens, density=True, alpha=0.5, bins='auto', label='neg_tokens')\n",
    "plt.xlim(0, 200)\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "train_val_df[train_val_df.pcl == 1].text.apply(len)\n",
    "train_val_df[train_val_df.pcl == 0].text.apply(len)\n",
    "plt.hist(train_val_df[train_val_df.pcl == 1].text.apply(len).values, density=True, alpha=0.5, bins='auto', label='pos_length')\n",
    "plt.hist(train_val_df[train_val_df.pcl == 0].text.apply(len).values, density=True, alpha=0.5, bins='auto', label='neg_length')\n",
    "plt.xlim(0, 1000)\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...\n",
      "To disable this warning, you can either:\n",
      "\t- Avoid using `tokenizers` before the fork if possible\n",
      "\t- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)\n",
      "huggingface/tokenizers: The current process just got forked, after parallelism has already been used. Disabling parallelism to avoid deadlocks...\n",
      "To disable this warning, you can either:\n",
      "\t- Avoid using `tokenizers` before the fork if possible\n",
      "\t- Explicitly set the environment variable TOKENIZERS_PARALLELISM=(true | false)\n"
     ]
    }
   ],
   "source": [
    "import nlpaug.augmenter.word as naw\n",
    "\n",
    "# can change language from any of the available models from https://huggingface.co/models?pipeline_tag=translation&sort=downloads&search=Helsinki-NLP\n",
    "back_translate_aug = naw.BackTranslationAug(from_model_name='Helsinki-NLP/opus-mt-en-zh', to_model_name='Helsinki-NLP/opus-mt-zh-en')    \n",
    "back_translated_text = back_translate_aug.augment(list(train_val_df[train_val_df.pcl == 1].text), num_thread=2)\n",
    "\n",
    "data = {'text': back_translated_text, 'pcl': np.ones(len(back_translated_text))}\n",
    "back_translated_df = pd.DataFrame(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>pcl</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>959</th>\n",
       "      <td>In the report , Madius added that the threat p...</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6184</th>\n",
       "      <td>Comprising two end terrace , two storey , hous...</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5491</th>\n",
       "      <td>IPOH : As Ramadan draws to a close , Muslims s...</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3134</th>\n",
       "      <td>KOTA KINABALU , May 21 ( Bernama ) -- A total ...</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>But in this one a pair of trends were bucked ,...</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>823</th>\n",
       "      <td>But Emil S ? nze , a local representative of t...</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3671</th>\n",
       "      <td>Social venture gives sanitary supplies for wom...</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2903</th>\n",
       "      <td>Pupils from poor families were given new shoes...</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3295</th>\n",
       "      <td>Tell us a little about your character Rishi in...</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2315</th>\n",
       "      <td>\" Everyone makes a mistake , no one 's perfect...</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>9169 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                   text  pcl\n",
       "959   In the report , Madius added that the threat p...  0.0\n",
       "6184  Comprising two end terrace , two storey , hous...  0.0\n",
       "5491  IPOH : As Ramadan draws to a close , Muslims s...  1.0\n",
       "3134  KOTA KINABALU , May 21 ( Bernama ) -- A total ...  0.0\n",
       "15    But in this one a pair of trends were bucked ,...  0.0\n",
       "...                                                 ...  ...\n",
       "823   But Emil S ? nze , a local representative of t...  0.0\n",
       "3671  Social venture gives sanitary supplies for wom...  0.0\n",
       "2903  Pupils from poor families were given new shoes...  0.0\n",
       "3295  Tell us a little about your character Rishi in...  0.0\n",
       "2315  \" Everyone makes a mistake , no one 's perfect...  1.0\n",
       "\n",
       "[9169 rows x 2 columns]"
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "augmented_df = pd.concat([train_val_df, back_translated_df])\n",
    "augmented_df.sample(frac=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "pos_tokens = CountVectorizer().fit_transform(augmented_df[augmented_df.pcl == 1].text.values)\n",
    "pos_num_tokens = np.sum(pos_tokens.todense(), axis=1).flat\n",
    "\n",
    "neg_tokens = CountVectorizer().fit_transform(augmented_df[augmented_df.pcl == 0].text.values)\n",
    "neg_num_tokens = np.sum(neg_tokens.todense(), axis=1).flat\n",
    "\n",
    "plt.hist(pos_num_tokens, density=True, alpha=0.5, bins='auto', label='pos_tokens')\n",
    "plt.hist(neg_num_tokens, density=True, alpha=0.5, bins='auto', label='neg_tokens')\n",
    "plt.xlim(0, 200)\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "augmented_df[augmented_df.pcl == 1].text.apply(len)\n",
    "augmented_df[augmented_df.pcl == 0].text.apply(len)\n",
    "plt.hist(augmented_df[augmented_df.pcl == 1].text.apply(len).values, density=True, alpha=0.5, bins='auto', label='pos_length')\n",
    "plt.hist(augmented_df[augmented_df.pcl == 0].text.apply(len).values, density=True, alpha=0.5, bins='auto', label='neg_length')\n",
    "plt.xlim(0, 1000)\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Apply Contextual Word Embedding Augmentation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [],
   "source": [
    "context_word_aug = naw.ContextualWordEmbsAug(action='insert')\n",
    "context_word_text = context_word_aug.augment(list(augmented_df[augmented_df.pcl == 1].text.values))\n",
    "\n",
    "data = {'text': context_word_text, 'pcl': np.ones(len(context_word_text))}\n",
    "context_word_df = pd.DataFrame(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [],
   "source": [
    "augmented_df = pd.concat([augmented_df, context_word_df])\n",
    "augmented_df = augmented_df.sample(frac=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {},
   "outputs": [],
   "source": [
    "augmented_df = augmented_df.reset_index()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Attempt Model on Augmented Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import DistilBertModel, DistilBertTokenizer\n",
    "import torch\n",
    "import torch.nn.functional as F\n",
    "from torch.utils.data import DataLoader\n",
    "from torch import cuda\n",
    "import numpy as np\n",
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "metadata": {},
   "outputs": [],
   "source": [
    "MAX_LEN = 512\n",
    "TRAIN_BATCH_SIZE = 8\n",
    "VALID_BATCH_SIZE = 2\n",
    "EPOCHS = 1\n",
    "LEARNING_RATE = 1e-05\n",
    "DEVICE = 'cuda' if cuda.is_available() else 'cpu'\n",
    "tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-cased')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import Dataset\n",
    "\n",
    "class Triage(Dataset):\n",
    "    def __init__(self, dataframe, tokenizer, max_len):\n",
    "        self.len = len(dataframe)\n",
    "        self.data = dataframe\n",
    "        self.tokenizer = tokenizer\n",
    "        self.max_len = max_len\n",
    "        \n",
    "    def __getitem__(self, index):\n",
    "        title = str(self.data.TITLE[index])\n",
    "        title = \" \".join(title.split())\n",
    "        inputs = self.tokenizer.encode_plus(\n",
    "            title,\n",
    "            None,\n",
    "            add_special_tokens=True,\n",
    "            max_length=self.max_len,\n",
    "            pad_to_max_length=True,\n",
    "            return_token_type_ids=True,\n",
    "            truncation=True\n",
    "        )\n",
    "        ids = inputs['input_ids']\n",
    "        mask = inputs['attention_mask']\n",
    "\n",
    "        return {\n",
    "            'ids': torch.tensor(ids, dtype=torch.long),\n",
    "            'mask': torch.tensor(mask, dtype=torch.long),\n",
    "            'targets': torch.tensor(self.data.ENCODE_CAT[index], dtype=torch.long)\n",
    "        } \n",
    "    \n",
    "    def __len__(self):\n",
    "        return self.len"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {},
   "outputs": [],
   "source": [
    "encode_dict = {} # dictionary of all unique output labels\n",
    "\n",
    "def encode_cat(x):\n",
    "    if x not in encode_dict.keys():\n",
    "        encode_dict[x]=len(encode_dict)\n",
    "    return encode_dict[x]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data = augmented_df[['text', 'pcl']].rename(columns={'text': \"TITLE\"})\n",
    "dev_data = test_df[['text', 'pcl']].rename(columns={'text': \"TITLE\"})\n",
    "\n",
    "train_data['ENCODE_CAT'] = train_data['pcl'].apply(lambda x: encode_cat(x))\n",
    "dev_data['ENCODE_CAT'] = dev_data['pcl'].apply(lambda x: encode_cat(x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_set = Triage(dataframe=train_data, tokenizer=tokenizer, max_len=MAX_LEN)\n",
    "dev_set = Triage(dataframe=dev_data, tokenizer=tokenizer, max_len=MAX_LEN)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_params = {'batch_size': TRAIN_BATCH_SIZE,\n",
    "                'shuffle': True,\n",
    "                'num_workers': 0\n",
    "                }\n",
    "\n",
    "test_params = {'batch_size': VALID_BATCH_SIZE,\n",
    "                'shuffle': True,\n",
    "                'num_workers': 0\n",
    "                }\n",
    "\n",
    "training_loader = DataLoader(train_set, **train_params)\n",
    "testing_loader = DataLoader(dev_set, **test_params)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 129,
   "metadata": {},
   "outputs": [],
   "source": [
    "class DistillBERTClass(torch.nn.Module):\n",
    "    def __init__(self):\n",
    "        super(DistillBERTClass, self).__init__()\n",
    "        self.l1 = DistilBertModel.from_pretrained(\"distilbert-base-uncased\")\n",
    "        self.pre_classifier = torch.nn.Linear(768, 768)\n",
    "        self.dropout = torch.nn.Dropout(0.3)\n",
    "        self.classifier = torch.nn.Linear(768, 1)\n",
    "\n",
    "    def forward(self, input_ids, attention_mask):\n",
    "        output_1 = self.l1(input_ids=input_ids, attention_mask=attention_mask)\n",
    "        hidden_state = output_1[0]\n",
    "        pooler = hidden_state[:, 0]\n",
    "        pooler = self.pre_classifier(pooler)\n",
    "        pooler = torch.nn.ReLU()(pooler)\n",
    "        pooler = self.dropout(pooler)\n",
    "        output = self.classifier(pooler)\n",
    "        return output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 130,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calcuate_accu(big_idx, targets):\n",
    "    n_correct = (big_idx==targets).sum().item()\n",
    "    return n_correct"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(epoch, model, loss_function, optimizer, device=DEVICE):\n",
    "    tr_loss = 0\n",
    "    n_correct = 0\n",
    "    nb_tr_steps = 0\n",
    "    nb_tr_examples = 0\n",
    "    model.train()\n",
    "    for _, data in tqdm(enumerate(training_loader, 0)):\n",
    "        ids = data['ids'].to(device, dtype = torch.long)\n",
    "        mask = data['mask'].to(device, dtype = torch.long)\n",
    "        targets = data['targets'].to(device, dtype = torch.float64)\n",
    "\n",
    "        outputs = model(ids, mask).flatten()\n",
    "        loss = loss_function(outputs, targets)\n",
    "\n",
    "        tr_loss += loss.item()\n",
    "        pred_val = torch.where(torch.nn.functional.sigmoid(outputs.data) >= 0.5, 1, 0)\n",
    "        # big_val, big_idx = torch.max(outputs.data, dim=1)\n",
    "        n_correct += calcuate_accu(pred_val, targets)\n",
    "\n",
    "        nb_tr_steps += 1\n",
    "        nb_tr_examples+=targets.size(0)\n",
    "        \n",
    "        if _%100==0:\n",
    "            loss_step = tr_loss/nb_tr_steps\n",
    "            accu_step = (n_correct*100)/nb_tr_examples \n",
    "            print(f\"Training Loss per 100 steps: {loss_step}\")\n",
    "            print(f\"Training Accuracy per 100 steps: {accu_step}\")\n",
    "\n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "    print(f'The Total Accuracy for Epoch {epoch}: {(n_correct*100)/nb_tr_examples}')\n",
    "    epoch_loss = tr_loss/nb_tr_steps\n",
    "    epoch_accu = (n_correct*100)/nb_tr_examples\n",
    "    print(f\"Training Loss Epoch: {epoch_loss}\")\n",
    "    print(f\"Training Accuracy Epoch: {epoch_accu}\")\n",
    "\n",
    "    return "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "metadata": {},
   "outputs": [],
   "source": [
    "def valid(model, testing_loader, loss_function, device=DEVICE):\n",
    "    tr_loss =0 \n",
    "    n_correct = 0\n",
    "    nb_tr_steps = 0\n",
    "    nb_tr_examples = 0\n",
    "    tp = 0\n",
    "    fp = 0\n",
    "    fn = 0\n",
    "    model.eval()\n",
    "    n_correct = 0; n_wrong = 0; total = 0\n",
    "    with torch.no_grad():\n",
    "        for _, data in enumerate(testing_loader, 0):\n",
    "            ids = data['ids'].to(device, dtype = torch.long)\n",
    "            mask = data['mask'].to(device, dtype = torch.long)\n",
    "            targets = data['targets'].to(device, dtype = torch.float64)\n",
    "            outputs = model(ids, mask).squeeze()\n",
    "            loss = loss_function(outputs.flatten(), targets)\n",
    "            tr_loss += loss.item()\n",
    "            pred_val = torch.where(torch.nn.functional.sigmoid(outputs.data) >= 0.5, 1, 0)\n",
    "            n_correct += calcuate_accu(pred_val, targets)\n",
    "            tp += torch.sum(torch.where(targets == 1, pred_val == targets, 0)).item()\n",
    "            fp += torch.sum(torch.where(targets == 0, pred_val != targets, 0)).item()\n",
    "            fn += torch.sum(torch.where(targets == 1, pred_val != targets, 0)).item()\n",
    "\n",
    "            nb_tr_steps += 1\n",
    "            nb_tr_examples+=targets.size(0)\n",
    "            \n",
    "            if _%100==0:\n",
    "                loss_step = tr_loss/nb_tr_steps\n",
    "                accu_step = (n_correct*100)/nb_tr_examples\n",
    "                print(f\"Validation Loss per 100 steps: {loss_step}\")\n",
    "                print(f\"Validation Accuracy per 100 steps: {accu_step}\")\n",
    "    epoch_loss = tr_loss/nb_tr_steps\n",
    "    epoch_accu = (n_correct*100)/nb_tr_examples\n",
    "    epoch_f1 = tp / (tp + 0.5 * (fp + fn))\n",
    "    print(f\"Validation Loss Epoch: {epoch_loss}\")\n",
    "    print(f\"Validation Accuracy Epoch: {epoch_accu}\")\n",
    "    print(f\"Validation F1 Epoch: {epoch_f1}\")\n",
    "\n",
    "    \n",
    "    return epoch_accu, epoch_f1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DistillBERTClass(\n",
       "  (l1): DistilBertModel(\n",
       "    (embeddings): Embeddings(\n",
       "      (word_embeddings): Embedding(30522, 768, padding_idx=0)\n",
       "      (position_embeddings): Embedding(512, 768)\n",
       "      (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
       "      (dropout): Dropout(p=0.1, inplace=False)\n",
       "    )\n",
       "    (transformer): Transformer(\n",
       "      (layer): ModuleList(\n",
       "        (0-5): 6 x TransformerBlock(\n",
       "          (attention): MultiHeadSelfAttention(\n",
       "            (dropout): Dropout(p=0.1, inplace=False)\n",
       "            (q_lin): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (k_lin): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (v_lin): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (out_lin): Linear(in_features=768, out_features=768, bias=True)\n",
       "          )\n",
       "          (sa_layer_norm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
       "          (ffn): FFN(\n",
       "            (dropout): Dropout(p=0.1, inplace=False)\n",
       "            (lin1): Linear(in_features=768, out_features=3072, bias=True)\n",
       "            (lin2): Linear(in_features=3072, out_features=768, bias=True)\n",
       "            (activation): GELUActivation()\n",
       "          )\n",
       "          (output_layer_norm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)\n",
       "        )\n",
       "      )\n",
       "    )\n",
       "  )\n",
       "  (pre_classifier): Linear(in_features=768, out_features=768, bias=True)\n",
       "  (dropout): Dropout(p=0.3, inplace=False)\n",
       "  (classifier): Linear(in_features=768, out_features=1, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 141,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "weighted_loss_model = DistillBERTClass()\n",
    "weighted_loss_model.to(DEVICE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "metadata": {},
   "outputs": [],
   "source": [
    "# weighting = sum(train_val_df.pcl == 0) / sum(train_val_df.pcl == 1) # add weighting to positive samples\n",
    "# loss_function = torch.nn.BCEWithLogitsLoss(pos_weight=torch.Tensor([weighting]))\n",
    "loss_function = torch.nn.BCEWithLogitsLoss()\n",
    "optimizer = torch.optim.Adam(params =  weighted_loss_model.parameters(), lr=LEARNING_RATE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "0it [00:00, ?it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.6713228831067681\n",
      "Training Accuracy per 100 steps: 75.0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100it [06:12,  4.18s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.5993281658450694\n",
      "Training Accuracy per 100 steps: 72.02970297029702\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "200it [12:58,  3.92s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.6109172639046652\n",
      "Training Accuracy per 100 steps: 70.58457711442786\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "300it [19:34,  3.92s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.6126772942005245\n",
      "Training Accuracy per 100 steps: 70.01661129568106\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "400it [26:16,  3.98s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.609610512585291\n",
      "Training Accuracy per 100 steps: 69.9501246882793\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "500it [34:33,  6.34s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.6010309594960794\n",
      "Training Accuracy per 100 steps: 69.81037924151697\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "600it [43:36,  5.17s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.5972487042410479\n",
      "Training Accuracy per 100 steps: 69.96672212978369\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "700it [53:25,  6.02s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.5910401170365526\n",
      "Training Accuracy per 100 steps: 70.38159771754636\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "800it [1:01:40,  4.47s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.581976653323346\n",
      "Training Accuracy per 100 steps: 71.08302122347067\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "900it [1:08:28,  3.86s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.5738829673169962\n",
      "Training Accuracy per 100 steps: 71.83684794672585\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1000it [1:14:56,  3.84s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.5653724996941759\n",
      "Training Accuracy per 100 steps: 72.42757242757243\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1100it [1:21:39,  4.64s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.5540667494200043\n",
      "Training Accuracy per 100 steps: 73.18346957311535\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1200it [1:28:39,  4.06s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.5451033315279147\n",
      "Training Accuracy per 100 steps: 73.68859283930058\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1300it [1:35:16,  3.87s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Loss per 100 steps: 0.5337297194937121\n",
      "Training Accuracy per 100 steps: 74.45234435049962\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1345it [1:38:13,  4.38s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The Total Accuracy for Epoch 0: 74.67695454122897\n",
      "Training Loss Epoch: 0.5308703692825026\n",
      "Training Accuracy Epoch: 74.67695454122897\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "for epoch in range(EPOCHS):\n",
    "    train(epoch, weighted_loss_model, loss_function, optimizer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation Loss per 100 steps: 0.2500517889857292\n",
      "Validation Accuracy per 100 steps: 100.0\n",
      "Validation Loss per 100 steps: 0.321276553104272\n",
      "Validation Accuracy per 100 steps: 88.61386138613861\n",
      "Validation Loss per 100 steps: 0.3140107339702139\n",
      "Validation Accuracy per 100 steps: 89.05472636815921\n",
      "Validation Loss per 100 steps: 0.3128758801326799\n",
      "Validation Accuracy per 100 steps: 89.53488372093024\n",
      "Validation Loss per 100 steps: 0.3231745918148994\n",
      "Validation Accuracy per 100 steps: 88.90274314214464\n",
      "Validation Loss per 100 steps: 0.33442290809875475\n",
      "Validation Accuracy per 100 steps: 88.62275449101796\n",
      "Validation Loss per 100 steps: 0.343310417777885\n",
      "Validation Accuracy per 100 steps: 87.93677204658901\n",
      "Validation Loss per 100 steps: 0.3345899531689503\n",
      "Validation Accuracy per 100 steps: 88.37375178316691\n",
      "Validation Loss per 100 steps: 0.3337548939387934\n",
      "Validation Accuracy per 100 steps: 88.5143570536829\n",
      "Validation Loss per 100 steps: 0.3319214258433943\n",
      "Validation Accuracy per 100 steps: 88.51276359600443\n",
      "Validation Loss per 100 steps: 0.33270010763740204\n",
      "Validation Accuracy per 100 steps: 88.56143856143856\n",
      "Validation Loss Epoch: 0.33088358462127254\n",
      "Validation Accuracy Epoch: 88.67654085045389\n",
      "Validation F1 Epoch: 0.3008849557522124\n",
      "Accuracy on test data = 88.68%\n",
      "F1 on test data = 0.30%\n"
     ]
    }
   ],
   "source": [
    "weighted_loss_acc, weighted_loss_f1 = valid(weighted_loss_model, testing_loader, loss_function)\n",
    "print(\"Accuracy on test data = %0.2f%%\" % weighted_loss_acc)\n",
    "print(\"F1 on test data = %0.2f%%\" % weighted_loss_f1)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "venv",
   "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": "3.10.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}