kan_diabetes

In this repo, I am trying to implement the Kolmogorov Arnold Networks (KAN) on a diabetes dataset.

Updates

This repo was updated on 5/28/2024. Changes were made to the branch to improve formula calculations.

Updates Made

• Implemented additional preprocessing steps for the diabetes dataset.
• Integrated symbolic regression to derive formulas. Improved the model plots before and after training. Added train_acc() and test_acc() functions to the model.train() function.
• Improved accuracy calculation and pruned model plotting.

Suggestions

• Consider implementing feature engineering for understanding the features present.
• One-Hot Encoding the categorical variables and building a correlation matrix.
• Train the model with the loss function formula.

Credits

The initial thought and inspiration for the model was by Chukwuebuka Ezeokeke (Gaus Octavio).

This project was updated and enhanced by Chris D.

For more details, visit the GitHub repository.