TorchKAN 0.1.0

An easy-to-use PyTorch implementation of the Kolmogorov Arnold Network

TorchKAN: Simplified KAN Model Evaluation

This project demonstrates the training, validation, and quantization of the KAN model using PyTorch with CUDA acceleration. The torchkan model evaluates performance on the MNIST dataset, providing insights into the capabilities of Generalized Additive Models (GAMs).

Project Status: Under Development

The KAN model has shown promising results amidst various GAMs since the 1980s. This implementation, inspired by various sources, achieves over 97% accuracy on the MNIST dataset. As this model is still under study, further exploration into its full capabilities is ongoing.

Prerequisites

Ensure you have the following installed on your system:

• Python (version 3.9 or higher)
• CUDA Toolkit (corresponding to your PyTorch installation's CUDA version)
• cuDNN (compatible with your installed CUDA Toolkit)

Installation

Tested on MacOS and Linux.

1. Clone the Repository

Clone the torchkan repository and set up the project environment:

git clone https://github.com/1ssb/torchkan.git
cd torchkan
pip install -r requirements.txt

2. Configure CUDA environment variables if they are not already set:

export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

3. Configure Weights & Biases (wandb)

To track experiments and model performance with wandb:

1. Set Up wandb Account:

• Sign up or log in at Weights & Biases.
• Find your API key in your account settings.

2. Initialize wandb in Your Project:

Before running the training script, initialize wandb:

wandb login

When prompted, enter your API key. This will link your script executions to your wandb account.

3. Make sure to change the Entity name in the mnist.py file to your username instead of 1ssb

This script will train the model, validate it, and log performance metrics using wandb.

Contact

For any inquiries or support, contact: [e-mail]

Contributions

Contributions are welcome. Please raise issues as necessary after commit "Fin.", scheduled end-June, 2024. The code is licensed under the MIT License.

References

• [0] Ziming Liu et al., "KAN: Kolmogorov-Arnold Networks", 2024, arXiv. https://arxiv.org/abs/2404.19756
• [1] https://github.com/KindXiaoming/pykan
• [2] https://github.com/Blealtan/efficient-kan