TensorFlow Model Support

[ombhojane]

Add support for TensorFlow models in our XAI package. Currently, we only support scikit-learn models.

Approach

• Create a wrapper class for TensorFlow models
• Implement prediction interface to match scikit-learn's
• Update feature importance calculation for TensorFlow models

Testing

• Test with a TensorFlow model
• Verify all visualizations work with TensorFlow models
• Check feature importance calculation

[Swish78]

I would like to request this issue to be assigned to me. However, I’m having difficulty locating where to add support for TensorFlow models in the XAI package. Could you please provide guidance on which files or modules I should focus on? Thank you!

[StunningShield4504]

Please assign this issue to me i would like to work upon this.My approach:1.Adding Tensorflow model 2.Check all visualisations 3.Do all necessary calculations

[Sakeebhasan123456]

hello @ombhojane i saw some problems in the model and i would like to work on it

Known Issues and Solutions

During the development of this project, several issues were encountered. Below are the problems along with their causes and the solutions implemented to address them.

1. SHAP Values Error

• Cause: The TreeExplainer from the SHAP library does not support the MLPClassifier (neural networks).
• Solution: Used DeepExplainer for neural networks instead of TreeExplainer. Alternatively, models not supported by TreeExplainer were excluded when calculating SHAP values. This ensured that the explainability analysis was accurate and applicable to each model type.

2. Low Test Accuracy

• Cause: The test accuracy was significantly lower than the training accuracy, indicating potential overfitting or an issue with model selection.
• Solution:
  • Ensured that the model used for prediction was the best-performing one by evaluating multiple models and selecting the most suitable.
  • Implemented proper cross-validation to assess the model's ability to generalize to unseen data.
  • Applied regularization techniques in the neural network to prevent overfitting.
  • Considered gathering more data or augmenting the existing dataset to improve model robustness.

3. UserWarning about Feature Names

• Cause: The model was fitted without feature names, but the prediction data included them, leading to a mismatch.
• Solution: Ensured that both the training and prediction datasets had consistent formats. Specifically, when fitting the model and making predictions, the data was converted to the same type (e.g., both as DataFrames with feature names).

[ombhojane]

Thank you all for your interest in this issue! This is an important enhancement that requires careful planning. Let me provide some guidance:

Project Structure

The main files you'll need to modify are:

1. core.py - Update XAIWrapper to handle TensorFlow models
2. model_interpretability.py - Adapt SHAP calculations for TF models
3. utils.py - Extend utility functions to support TF models

Requirements

1. Experience with TensorFlow and scikit-learn
2. Understanding of model interpretability techniques
3. Familiarity with Python testing frameworks

Known Challenges

1. SHAP values calculation differs for deep learning models
2. Visualization adaptations may be needed
3. Performance considerations for large models

Would you like to share your experience with TensorFlow and model interpretability? This will help me assign the issue to the best-suited contributor.

[ombhojane]

okayy @Sakeebhasan123456 go ahead

[Sakeebhasan123456]

Hi @ombhojane i am trying to solve this problem please check also add a pull request

[garvender004]

Hello @ombhojane I would like to contribute to this project.