Explainable AI (XAI)

Introduction

This repository contains explainable AI pipelines and functions that provide insightful visualizations and explanations for models across various domains, including computer vision and financial applications.

The tools and algorithms offered in this repository aim to enhance transparency in AI models by identifying and explaining feature contributions, uncovering patterns, and providing a deeper understanding of model behavior.

How to Run the Applications

Individual Apps

You can run the vision-focused or banking-focused apps independently:

1. Banking Application:

   streamlit run xai_banking/app.py

2. Vision Application:

   streamlit run main_page.py

Multipage App

If you want to run a combined app for both vision and banking functionalities, use the multipage application: bash streamlit run multi_app/main.py

The multipage app provides a single interface to select between Banking Pilot and Fmake Pilot, dynamically loading the respective application.

Environment Setup

To ensure a smooth setup and execution of the repository, follow these steps:

1. Clone the Repository Clone this repository to your local machine:

   git clone <repository-url>
   cd <repository-name> 

2. Set Up Conda Environment Create a Conda environment with all necessary dependencies:

Create the environment: bash conda create --name xai_env python=3.11 -y Activate the environment: bash conda activate xai_env Install dependencies: bash pip install -r requirements.txt

Build and run Docker

1. To build and run the Streamlit application using Docker, follow these steps:

   Build the Docker Image: bash docker build -t xai-app . -f streamlit.Dockerfile

   Run the Docker Container: bash docker run -d -p 8501:8501 --name xai-container --env-file .env xai-app

2. To build and run the FastAPI service using Docker, follow these steps:

   Build the Docker Image: bash docker build -t xai-api -f api.Dockerfile .

   Build the Docker Image: bash docker run -d -p 8000:8000 --name api-container --env-file .env xai-api

3. Creating the .env File

   Both the Streamlit app and the FastAPI service rely on environment variables defined in a .env file. Below is an example of how to create the .env file:

   # Paths for Banking Models and Data
   BANKING_MODEL_PATH=xai_banking/utils/random_forest_model.pkl
   BANKING_DATA_PATH=xai_banking/utils/synthetic_dataset.csv
   
   # Paths for Vision Models and Data
   VISION_MODEL_PATH=xaivision/model.onnx
   VISION_DATA_PATH=xaivision/data1.h5
   
   # API Key for FastAPI
   API_KEYS=["Your preference API key"]
   
   # Token for Streamlit Authentication
   STREAMLIT_TOKEN="Your preference Streamlit token"

General Use Tools

The general use tools that are currently offered in the repository are listed in the table below.

Tool	Description	Applicable Domain
Model Details	Visualize the model architecture with details about total parameters and number of operations.	Vision, Banking
Sample Details	Display input with the estimated output of the model.	Vision, Banking
Convolution Layers Activation	Display the values of the convolutional layers stacked together.	Vision
Independent Components	Split the input into independent components to identify patterns.	Vision
Preprocessed Data Overview	Visualize and inspect preprocessed data, such as encoded features or transformed datasets.	Banking
Feature Contributions	Analyze the impact of specific features (e.g., categorical or numerical) on model predictions.	Banking

Algorithms

The XAI algorithms currently offered in the repository are listed in the table below:

Tool	Description	Applicable Domain
Model Details	Visualize the model architecture with details about total parameters and number of operations.	Vision, Banking
Sample Details	Display input with the estimated output of the model.	Vision, Banking
Convolution Layers Activation	Display the values of the convolutional layers stacked together.	Vision
Independent Components	Split the input into independent components to identify patterns.	Vision
Preprocessed Data Overview	Visualize and inspect preprocessed data, such as encoded features or transformed datasets.	Banking
Feature Contributions	Analyze the impact of specific features (e.g., categorical or numerical) on model predictions.	Banking

Banking-Specific Features

The repository now includes additional features and tools tailored for financial and banking applications:

Tools

Tool	Description
Lime Explanation Viewer	Visualize feature contributions for individual predictions using LIME.
SHAP Explanation Viewer	Display global and local feature contributions using SHAP visualizations.
Top Features Viewer	Highlight the top contributing features for a given prediction.
Data Preprocessing Tools	Preprocess and encode financial datasets for explainability tasks.

Visualizations

Visualization	Description
Bar Chart (LIME)	Display the contribution of each feature for an individual prediction using LIME.
SHAP Summary Plot	Visualize the global feature importance and contribution distributions.
SHAP Waterfall Plot	Show the breakdown of individual feature contributions for a specific instance.
SHAP Heatmap Plot	Highlight patterns and clusters of feature contributions across multiple data points.
SHAP Beeswarm Plot	Combine feature importance and distribution across data points for global insights.

Applications

This repository supports two primary domains:

1. Vision Models:

• Explainability for image classification models.
• Focus on pixel-level contributions and visual feature activations.
• Tools: Vision SHAP, Integrated Gradients, DeepLift.

2. Banking Models:

• Explainability for financial and tabular data models.
• Feature-level analysis for predictions, such as fraud detection and credit scoring.
• Tools: LIME, SHAP, Preprocessed Data Overview, Feature Contributions.

How to Use

1. Select the Application:

• For vision tasks, use tools like Vision SHAP and Integrated Gradients.
• For banking tasks, use tools like LIME and SHAP Explanation Viewers.

2. Visualize and Interpret:

• Explore detailed visualizations, such as SHAP Summary and Waterfall plots for banking models, or pixel-level explanations for vision models.

3. Extend the Framework:

Add custom explainability tools or algorithms for other domains as needed.