chapter08_xai.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "4b7f9744",
   "metadata": {},
   "source": [
    "# Chapter 08 - Explainable AI (XAI) techniques \n",
    "\n",
    "Example of how attention visualization, saliency maps, and natural language explanations can be generated simply using Python"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "ef40fcba",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Import required libraries for model, tokenizer, and visualization\n",
    "import torch\n",
    "from transformers import BertTokenizer, BertForSequenceClassification\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "a1e56c31",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Some weights of BertForSequenceClassification were not initialized from the model checkpoint at bert-base-uncased and are newly initialized: ['classifier.bias', 'classifier.weight']\n",
      "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
     ]
    }
   ],
   "source": [
    "# Setup the model and tokenizer\n",
    "model_name = 'bert-base-uncased'  # Specify pre-trained model name\n",
    "tokenizer = BertTokenizer.from_pretrained(model_name)  # Load tokenizer for model\n",
    "model = BertForSequenceClassification.from_pretrained(model_name, num_labels=2)  # Load BERT model for sequence classification with 2 labels\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "0da15a7a",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Tokenize input text\n",
    "travel_query = \"What are the best family-friendly travel destinations in Europe?\"  # Example travel-related query\n",
    "inputs = tokenizer(travel_query, return_tensors=\"pt\", truncation=True, padding=True)  # Tokenize input query with padding and truncation\n",
    "input_ids = inputs['input_ids']  # Extract token IDs for the input text\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "2608836b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to get attention scores from the model\n",
    "def get_attention_scores(model, inputs):\n",
    "    # Run model with input and extract attention scores\n",
    "    outputs = model(**inputs, output_attentions=True)\n",
    "    return outputs.attentions  # Return attention scores from all layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "a095cb32",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to visualize attention scores\n",
    "def visualize_attention(attention_scores, tokens):\n",
    "    sns.set(style='whitegrid')  # Set seaborn style for heatmap\n",
    "    fig, ax = plt.subplots(figsize=(8, 8))  # Create a figure for plotting\n",
    "    attention_layer = attention_scores[-1][0]  # Extract the last attention layer's scores\n",
    "    attention_weights = attention_layer[0].detach().numpy()  # Convert attention scores to numpy array for plotting\n",
    "\n",
    "    # Plot the attention heatmap\n",
    "    sns.heatmap(attention_weights, xticklabels=tokens, yticklabels=tokens, cmap=\"viridis\", ax=ax)\n",
    "    plt.title(\"Attention Visualization\")  # Title for the plot\n",
    "    plt.show()  # Display the plot\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "12f2f13e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x800 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Get attention scores from the model\n",
    "attention_scores = get_attention_scores(model, inputs)\n",
    "\n",
    "# Decode the input IDs to get human-readable tokens\n",
    "tokens = tokenizer.convert_ids_to_tokens(input_ids[0])\n",
    "\n",
    "# Visualize the attention scores for the tokens in the input\n",
    "visualize_attention(attention_scores, tokens)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0238c484",
   "metadata": {},
   "source": [
    "## Saliency Maps"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "7adf3c8a",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Import necessary library for saliency map visualization\n",
    "from captum.attr import Saliency"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "eaf893cc",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Saliency Map Visualization function to visualize the importance of each token\n",
    "def visualize_saliency(sentence):\n",
    "    # Step 1: Tokenize the input sentence\n",
    "    # Use the BERT tokenizer to convert the input sentence into token IDs and attention masks.\n",
    "    # Attention mask ensures the padding tokens are ignored during processing.\n",
    "    inputs = tokenizer(sentence, return_tensors='pt', truncation=True, padding=True)\n",
    "    input_ids = inputs['input_ids']\n",
    "    attention_mask = inputs['attention_mask']\n",
    "\n",
    "    # Step 2: Ensure the model is in evaluation mode\n",
    "    # In evaluation mode, the model will not perform backpropagation (no parameter updates),\n",
    "    # which is important for inference and visualizations.\n",
    "    model.eval()\n",
    "    \n",
    "    # Step 3: Get embeddings and enable gradient tracking\n",
    "    # Get the token embeddings from the model's input embedding layer and enable gradient tracking.\n",
    "    # This allows the saliency map to be computed based on the gradients of the embeddings.\n",
    "    embeddings = model.get_input_embeddings()(input_ids).requires_grad_()\n",
    "\n",
    "    # Step 4: Define a custom forward function for the model\n",
    "    # The forward function will process the embeddings directly and return the model's logits (predictions).\n",
    "    # We bypass the usual model inputs to directly feed the embeddings for saliency calculation.\n",
    "    def forward_with_logits(embeddings):\n",
    "        return model(inputs_embeds=embeddings, attention_mask=attention_mask).logits\n",
    "\n",
    "    # Step 5: Initialize Saliency and compute the saliency scores\n",
    "    # Saliency scores measure the contribution of each token to the model's prediction.\n",
    "    saliency = Saliency(forward_with_logits)\n",
    "    \n",
    "    # Get the saliency scores for the input embeddings with respect to the positive class (target=1)\n",
    "    saliency_scores = saliency.attribute(embeddings, target=1)\n",
    "\n",
    "    # Step 6: Convert token IDs back to human-readable tokens\n",
    "    # This step helps us map the token IDs back into actual words for better interpretability.\n",
    "    tokens = tokenizer.convert_ids_to_tokens(input_ids[0])\n",
    "    \n",
    "    # Step 7: Aggregate the saliency scores for visualization\n",
    "    # Summing the saliency scores across the token's embedding dimension to get a single score for each token.\n",
    "    # This helps in visualizing the importance of each token in a 2D plot.\n",
    "    saliency_scores = saliency_scores.sum(dim=2).squeeze()\n",
    "\n",
    "    # Step 8: Visualize the saliency map\n",
    "    # Create a bar chart where each token's importance score is plotted, allowing us to see which tokens\n",
    "    # are most important for the model's decision. Tokens on the x-axis and saliency scores on the y-axis.\n",
    "    plt.bar(range(len(tokens)), saliency_scores.detach().numpy(), tick_label=tokens, color='teal')\n",
    "    \n",
    "    # Rotate x-axis labels for better readability of tokens\n",
    "    plt.xticks(rotation=45)\n",
    "    \n",
    "    # Add a title to the plot\n",
    "    plt.title(\"Saliency Map\")\n",
    "    \n",
    "    # Display the plot\n",
    "    plt.show"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "13084492",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Example input sentence\n",
    "travel_query = \"What are the best family-friendly travel destinations in Europe?\"\n",
    "\n",
    "# Example sentence to visualize saliency for\n",
    "visualize_saliency(travel_query)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5549796e",
   "metadata": {},
   "source": [
    "## Natural Language Explanations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "fcb1e5cb",
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install openai --quiet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "c5c0c434",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Import the OpenAI library and os for accessing environment variables\n",
    "from openai import OpenAI\n",
    "import os"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "17db0d26",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Set the OpenAI API key directly in the notebook\n",
    "os.environ[\"OPENAI_API_KEY\"] = \"YOUR-KEY\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "644ebf68",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Explanation:\n",
      "The question 'What are the best family-friendly travel destinations in Europe?' is significant for several reasons in the context of travel:\n",
      "\n",
      "1. **Diverse Needs of Families**: Families traveling with children have unique requirements compared to solo travelers or couples. This includes considerations for safety, availability of kid-friendly activities, and accommodations that cater to families. Identifying the best destinations helps families plan trips that meet these specific needs.\n",
      "\n",
      "2. **Quality Time and Bonding**: Family vacations provide essential opportunities for bonding and creating lasting memories. Finding destinations that offer activities and experiences suitable for all ages can enhance family dynamics and improve relationships.\n",
      "\n",
      "3. **Educational Opportunities**: Europe is rich in history, culture, and art. Family-friendly travel destinations often incorporate educational components, allowing children to learn while having fun. This can enrich children's understanding of the world, making travel both enjoyable and informative.\n",
      "\n",
      "4. **Health and Safety**: Choosing the right destination can also be crucial for health and safety. Certain locations may have better healthcare facilities, less crowded tourist areas, or environments that are more suitable for children, thus easing parents' worries.\n",
      "\n",
      "5. **Budget Considerations**: Families often travel on a budget. Knowing which destinations are family-friendly can help parents find affordable options for accommodations, meals, and activities that provide good value for money.\n",
      "\n",
      "6. **Cultural Exposure**: Traveling to Europe can expose families to different cultures, traditions, and languages. Family-friendly destinations often facilitate these experiences in a manner that's engaging and accessible for children.\n",
      "\n",
      "7. **Travel Logistics**: Family travel often involves more logistical planning, including transportation and accessibility. Knowing which destinations are easier to navigate with children can simplify the travel experience and reduce stress.\n",
      "\n",
      "8. **Encouragement for Travel**: By highlighting family-friendly options, more families may be encouraged to travel. This can lead to increased cultural exchange and a broader appreciation for global diversity among younger generations.\n",
      "\n",
      "Overall, understanding the best family-friendly travel destinations in Europe allows families to make informed choices that enhance their travel experiences, contributing positively to their overall well-being and enjoyment.\n"
     ]
    }
   ],
   "source": [
    "# Function to generate an explanation using OpenAI\n",
    "def generate_explanation_with_openai(text):\n",
    "    # Initialize the OpenAI client with the API key stored in the environment variables\n",
    "    client = OpenAI(api_key=os.environ[\"OPENAI_API_KEY\"])\n",
    "\n",
    "    # Use the OpenAI client to make a request for generating a completion (explanation)\n",
    "    response = client.chat.completions.create(\n",
    "        model=\"gpt-4o-mini\",  # Specify the model to be used for generation\n",
    "        messages=[  # Provide the context for the model\n",
    "            {\"role\": \"system\", \"content\": \"You are an explainability assistant.\"},  # System message setting the assistant's role\n",
    "            {\"role\": \"user\", \"content\": f\"Explain why '{text}' is important in the context of travel.\"}  # User query with the input text\n",
    "        ]\n",
    "    )\n",
    "\n",
    "    # Extract the generated explanation from the response object\n",
    "    explanation = response.choices[0].message.content\n",
    "\n",
    "    # Return the explanation text\n",
    "    return explanation\n",
    "\n",
    "# Example usage: calling the function with a sample input\n",
    "explanation = generate_explanation_with_openai(travel_query)\n",
    "\n",
    "# Output the generated explanation\n",
    "print(f\"Explanation:\\n{explanation}\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7c972cfc",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "conda_pytorch_p310",
   "language": "python",
   "name": "conda_pytorch_p310"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}