KAN_for_GP

This repository contains the implementation of various models evaluated in the study "Exploring Kolmogorov-Arnold Networks for High-Dimensional Genomic Prediction".

📂 Project Structure

• encoder.py: Implements all comparison models used in the paper.

  • classDNNGP: CNN.
  • MLP: Multi-layer perceptron.
  • KAN: Kolmogorov–Arnold Network.
  • KAN_MLP: A hybrid model where genomic data is first processed by KAN, and its output is then passed to an MLP for final prediction.
  • LinformerEncoderLayer: Linformer-based encoder.

• fit_torch.py: Entry point to run model training and evaluation with selected hyperparameters.

• torch_get_params.py: The ray_hyperparam_search function defines the hyperparameter search spaces for all models.

• torch_shared_functions.py: Contains utility functions shared across model training and evaluation routines.

⚙️ Environment

To recreate the environment, install dependencies listed in requirements.txt.

📝 Notes

• Data preprocessing and raw genomic datasets are not included. The original datasets can be accessed from the CropGS-Hub repository.

📄 License

This project is intended for academic use only.