demo_user_input.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# BERT-based Question Answering Demo\n",
    "\n",
    "This project demonstrates how to use a fine-tuned BERT model for the **Question Answering (QA)** task. It leverages the Hugging Face `transformers` library with TensorFlow to load a pre-trained model and tokenizer for answering questions based on a given context.\n",
    "\n",
    "## Project Structure\n",
    "- **`trained_model/`**: Directory containing the fine-tuned BERT model for question answering.\n",
    "- **`data/`**: Directory containing the `dataset.json` file with context data for answering questions.\n",
    "- **`script.py`**: The main Python script that performs question answering.\n",
    "\n",
    "## Requirements\n",
    "- TensorFlow\n",
    "- Hugging Face Transformers library\n",
    "- JSON dataset containing contexts for answering questions\n",
    "\n",
    "## How It Works\n",
    "\n",
    "1. **Model and Tokenizer Initialization**:\n",
    "   - The model is loaded from the specified `trained_model` directory using the `TFBertForQuestionAnswering` class.\n",
    "   - The tokenizer is loaded from the same directory using `BertTokenizerFast`.\n",
    "\n",
    "2. **Answering a Question**:\n",
    "   - The `QAInference` class defines the process for taking a `context` (a passage of text) and a `question`, and using the model to predict the answer.\n",
    "   - The input question and context are tokenized using the tokenizer.\n",
    "   - The model predicts the start and end positions of the answer in the context.\n",
    "   - The predicted span of text is decoded back into a readable string and returned as the answer.\n",
    "\n",
    "3. **User Interaction**:\n",
    "   - When the script runs, users are prompted to enter a question.\n",
    "   - The context for the question is retrieved from the first entry in the `dataset.json` file (which you can modify as needed).\n",
    "   - The script outputs the question along with the model's predicted answer.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Question: What does natural language processing enable?\n",
      "Predicted Answer: building systems that learn from data\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "from transformers import TFBertForQuestionAnswering, BertTokenizerFast\n",
    "import json\n",
    "import os\n",
    "\n",
    "# Directory stuff\n",
    "workspace_dir = os.path.abspath(os.path.join(os.getcwd()))\n",
    "model_path = os.path.join(workspace_dir, 'trained_model')\n",
    "dataset_path = os.path.join(workspace_dir, 'data', 'dataset.json')\n",
    "\n",
    "class QAInference:\n",
    "    def __init__(self, model_path):\n",
    "        \"\"\"\n",
    "        Initializes the model and tokenizer for inference.\n",
    "        \"\"\"\n",
    "        # Load the fine-tuned model and tokenizer\n",
    "        self.model = TFBertForQuestionAnswering.from_pretrained(model_path)\n",
    "        self.tokenizer = BertTokenizerFast.from_pretrained(model_path)\n",
    "\n",
    "    def answer_question(self, context, question):\n",
    "        \"\"\"\n",
    "        Given a context and a question, return the answer predicted by the model.\n",
    "        \"\"\"\n",
    "        # Tokenize the input question and context\n",
    "        inputs = self.tokenizer(question, context, return_tensors=\"tf\", truncation=True, padding=True)\n",
    "        \n",
    "        # Get model outputs\n",
    "        outputs = self.model(inputs)\n",
    "        \n",
    "        # Find the start and end logits of the predicted answer\n",
    "        start_scores = outputs.start_logits\n",
    "        end_scores = outputs.end_logits\n",
    "        \n",
    "        # Get the start and end positions of the answer\n",
    "        start_idx = tf.argmax(start_scores, axis=1).numpy()[0]\n",
    "        end_idx = tf.argmax(end_scores, axis=1).numpy()[0]\n",
    "        \n",
    "        # Extract the answer tokens and decode to string\n",
    "        answer_tokens = inputs[\"input_ids\"][0][start_idx:end_idx + 1]\n",
    "        answer = self.tokenizer.decode(answer_tokens, skip_special_tokens=True)\n",
    "        \n",
    "        return answer\n",
    "\n",
    "def main():\n",
    "    # Load dataset from JSON (just for the context)\n",
    "    with open(dataset_path, 'r') as file:\n",
    "        dataset = json.load(file)\n",
    "\n",
    "    # Initialize the QA model\n",
    "    qa_model = QAInference(model_path)\n",
    "\n",
    "    # Allow user to input a question\n",
    "    question = input(\"Enter your question: \")\n",
    "\n",
    "    # Use the first context from the dataset (or modify as needed)\n",
    "    context = dataset[0][\"context\"]\n",
    "\n",
    "    # Get the predicted answer for the given question and context\n",
    "    predicted_answer = qa_model.answer_question(context, question)\n",
    "\n",
    "    print(\"\\nQuestion:\", question)\n",
    "    print(\"Predicted Answer:\", predicted_answer)\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "    main()\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": "3.12.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}