{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\Admin\\AppData\\Local\\Programs\\Python\\Python311\\Lib\\site-packages\\transformers\\optimization.py:429: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use the PyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning\n",
      "  warnings.warn(\n",
      "Epoch 1/3: 100%|██████████| 1293/1293 [35:14<00:00,  1.64s/batch, Loss=0.000204]\n",
      "Epoch 2/3: 100%|██████████| 1293/1293 [35:04<00:00,  1.63s/batch, Loss=0.000154]\n",
      "Epoch 3/3: 100%|██████████| 1293/1293 [34:39<00:00,  1.61s/batch, Loss=7.12e-5] \n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "from transformers import AutoTokenizer, AutoModelForMaskedLM, AdamW\n",
    "import pandas as pd\n",
    "from tqdm import tqdm\n",
    "\n",
    "class GrammarCorrectionDataset(Dataset):\n",
    "    def __init__(self, sentences, corrected_sentences, tokenizer, max_length=128):\n",
    "        self.sentences = sentences\n",
    "        self.corrected_sentences = corrected_sentences\n",
    "        self.tokenizer = tokenizer\n",
    "        self.max_length = max_length\n",
    "\n",
    "    def __len__(self):\n",
    "        return len(self.sentences)\n",
    "    \n",
    "    def __getitem__(self, idx):\n",
    "        input_sentence = self.sentences[idx]\n",
    "        corrected_sentence = self.corrected_sentences[idx]\n",
    "\n",
    "        # Tokenize the input and corrected sentences separately\n",
    "        inputs = self.tokenizer(\n",
    "            [input_sentence],  # Pass as a list\n",
    "            [corrected_sentence],  # Pass as a list\n",
    "            return_tensors=\"pt\",\n",
    "            padding=\"max_length\",\n",
    "            truncation=True,\n",
    "            max_length=self.max_length\n",
    "        )\n",
    "\n",
    "        return {\n",
    "            \"input_ids\": inputs[\"input_ids\"].flatten(),\n",
    "            \"attention_mask\": inputs[\"attention_mask\"].flatten(),\n",
    "            \"labels\": inputs[\"input_ids\"].flatten()  # Use input_ids as labels for MLM\n",
    "        }\n",
    "\n",
    "\n",
    "\n",
    "def pad_collate(batch):\n",
    "    # Find the length of the longest sentence in the batch\n",
    "    max_len = max(len(batch_item[\"input_ids\"]) for batch_item in batch)\n",
    "    \n",
    "    # Pad each input to the length of the longest sentence in the batch\n",
    "    for batch_item in batch:\n",
    "        input_ids = batch_item[\"input_ids\"]\n",
    "        attention_mask = batch_item[\"attention_mask\"]\n",
    "        labels = batch_item[\"labels\"]\n",
    "        \n",
    "        padded_input_ids = torch.nn.functional.pad(input_ids, (0, max_len - len(input_ids)), value=tokenizer.pad_token_id)\n",
    "        padded_attention_mask = torch.nn.functional.pad(attention_mask, (0, max_len - len(attention_mask)), value=0)\n",
    "        padded_labels = torch.nn.functional.pad(labels, (0, max_len - len(labels)), value=tokenizer.pad_token_id)\n",
    "        \n",
    "        batch_item[\"input_ids\"] = padded_input_ids\n",
    "        batch_item[\"attention_mask\"] = padded_attention_mask\n",
    "        batch_item[\"labels\"] = padded_labels\n",
    "        \n",
    "    return {\n",
    "        \"input_ids\": torch.stack([batch_item[\"input_ids\"] for batch_item in batch]),\n",
    "        \"attention_mask\": torch.stack([batch_item[\"attention_mask\"] for batch_item in batch]),\n",
    "        \"labels\": torch.stack([batch_item[\"labels\"] for batch_item in batch])\n",
    "    }\n",
    "\n",
    "\n",
    "data = pd.read_csv(r'D:\\Thesis\\test_bert_data.csv')\n",
    "data = data.dropna()\n",
    "sentences = [str(i) for i in data['wrong'].values] # Convert Series to list\n",
    "corrected_sentences = [str(i) for i in data['right1'].values]\n",
    "\n",
    "# Initialize tokenizer and model\n",
    "tokenizer = AutoTokenizer.from_pretrained(\"jcblaise/roberta-tagalog-base\")\n",
    "model = AutoModelForMaskedLM.from_pretrained(\"jcblaise/roberta-tagalog-base\")\n",
    "\n",
    "# Create dataset and dataloader\n",
    "dataset = GrammarCorrectionDataset(sentences, corrected_sentences, tokenizer)\n",
    "dataloader = DataLoader(dataset, batch_size=4, shuffle=True, collate_fn=pad_collate)\n",
    "\n",
    "\n",
    "# Define optimizer and loss function\n",
    "optimizer = AdamW(model.parameters(), lr=5e-5)\n",
    "\n",
    "num_epochs = 3\n",
    "\n",
    "model.train()\n",
    "for epoch in range(num_epochs):\n",
    "    tqdm_dataloader = tqdm(dataloader, desc=f\"Epoch {epoch + 1}/{num_epochs}\", unit=\"batch\")\n",
    "    for batch in tqdm_dataloader:\n",
    "        input_ids = batch[\"input_ids\"]\n",
    "        attention_mask = batch[\"attention_mask\"]\n",
    "        labels = batch[\"labels\"]\n",
    "\n",
    "        outputs = model(input_ids=input_ids, attention_mask=attention_mask, labels=labels)\n",
    "        loss = outputs.loss\n",
    "\n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        # Update tqdm progress bar description with current loss\n",
    "        tqdm_dataloader.set_postfix({\"Loss\": loss.item()})\n",
    "\n",
    "    # Close the tqdm progress bar for the epoch\n",
    "    tqdm_dataloader.close()\n",
    "\n",
    "# Save the fine-tuned model\n",
    "#model.save_pretrained(\"fine_tuned_model\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save_pretrained(\"fine_tuned_model\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Asking to truncate to max_length but no maximum length is provided and the model has no predefined maximum length. Default to no truncation.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Original: Takbo <mask> takbo ang mga bata sa labas\n",
      "Corrected: <s>Takbo ng takbo ang mga bata sa labas</s>\n",
      "\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "from transformers import AutoTokenizer, AutoModelForMaskedLM\n",
    "\n",
    "# Load the fine-tuned model\n",
    "model = AutoModelForMaskedLM.from_pretrained(\"fine_tuned_model\")\n",
    "\n",
    "# Initialize tokenizer\n",
    "tokenizer = AutoTokenizer.from_pretrained(\"fine_tuned_model\")\n",
    "\n",
    "# Example new data\n",
    "new_data = [\n",
    "    \"Takbo <mask> takbo ang mga bata sa labas\"\n",
    "]\n",
    "\n",
    "# Tokenize the new data\n",
    "tokenized_data = tokenizer(new_data, return_tensors=\"pt\", padding=True, truncation=True)\n",
    "\n",
    "# Pass the tokenized data through the model to get predictions\n",
    "with torch.no_grad():\n",
    "    outputs = model(**tokenized_data)\n",
    "\n",
    "# Decode the predicted token IDs back to text\n",
    "predicted_texts = tokenizer.batch_decode(outputs.logits.argmax(dim=-1))\n",
    "\n",
    "# Print the original sentences and their corrected versions\n",
    "for original, corrected in zip(new_data, predicted_texts):\n",
    "    print(f\"Original: {original}\")\n",
    "    print(f\"Corrected: {corrected}\")\n",
    "    print()\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}