Ratheshan03 / ExplainableFL

ExplainableFL is a PIP Python package designed to bring explainability to Federated Learning models using SHAP values. It provides easy-to-use methods to visualize the impact of model features and privacy mechanisms on model performance.

ExplainableFL

ExplainableFL is a Python package designed to bring explainability to Federated Learning models using SHAP values. It provides easy-to-use methods to visualize the impact of model features and privacy mechanisms on model performance.

Installation

To install ExplainableFL, run the following command:

pip install -i https://test.pypi.org/simple/ explainablefl

Ensure you have the necessary prerequisites installed, including Python 3.6+ and pip

Usage

Here's a quick example to get you started:

from torch.utils.data import DataLoader, TensorDataset
from explainablefl import FederatedXAI

# Example model
model = torch.nn.Linear(10, 2)

# Setup DataLoader
x = torch.randn(100, 10)
y = torch.randint(0, 2, (100,))
dataset = TensorDataset(x, y)
data_loader = DataLoader(dataset, batch_size=10)

# Initialize FederatedXAI
federated_xai = FederatedXAI(device=torch.device('cpu'), global_model=model, data_loader=data_loader)

# Use the library to explain the client model
shap_plot_buf, _ = federated_xai.explain_client_model(model)

Features

• Client Model Explanation: Visualize how individual features influence model predictions.
• Global Model Explanation: Generate SHAP explanations and confusion matrices.
• Aggregation Impact: Assess the effect of model aggregation in federated settings.
• Privacy Impact Analysis: Understand the influence of differential privacy mechanisms.

Contributing

Contributions are welcome! Please feel free to submit pull requests or open issues to improve the documentation, code quality, or add new features.

Next updates

• Extended functionality to support additional frameworks beyond PyTorch.
• Modify explainable functions to generate more interactive visualizations.
• Improve error handling.
• Testing CI/CD pipelines.

License

This project is licensed under the MIT License - see the LICENSE file for details.