Skip to content

Latest commit

 

History

History
34 lines (25 loc) · 1.91 KB

README.md

File metadata and controls

34 lines (25 loc) · 1.91 KB

SpIN Jax

This repository provides a jax implementation of Spectral Inference Networks: Unifying Deep and Spectral Learning. It is not the authors official tensorflow implementation which you can find here, but the results are very similar, and run thanks to jax's jit even a bit faster. However, at the moment it only supports Hamiltonian systems. If you are interested in general kernels you will need to add some small extra modifications.

Additionally, we provide a deep dive with a breakdown of the code, some extra explanations and additional experiments and visualizations in this colab notebook.

Open In Colab

Installation

You will need jax, tensorflow (for checkpointing) and flax (for the neural networks) along with a few other requirements. To get started run

conda create --name spin --file requirements.txt
conda activate spin

Usage

To run the training of SpIN for the wave functions of the hydrogen atoms just run

python train_spin.py

The hyperparameter are found and can be modified within the train_spin.py file. For now only system='laplacian' and system='hydrogen' are supported. If you want to give it a try for different Hamiltonian system you can add the corresponding potentials in physics.py, and increase n_space_dimension for multi particle systems accordingly.

Results

The following shows the training evolution for the first 4 eigenfunctions of the hydrogen atom.

Sample training