2-valued-neural-networks

Source code for the paper "Search for Z/2 eigenfunctions on the sphere using machine learning"

In the folder /models you can find the last two versions of our source code. In the folder /web you can view the statistics of our most interesting runs. It also has some 3d renders of our Z/2 eigenfunction.

Quick links

Statistics of our runs: https://wasalm.github.io/2-valued-neural-networks/ 3d Render of in the

• Tetrehedral case: https://wasalm.github.io/2-valued-neural-networks/sphere.html?tetrahedra
• Squashed tetrahedral case: https://wasalm.github.io/2-valued-neural-networks/sphere.html?squashed-tetrahedra
• Cubic case: https://wasalm.github.io/2-valued-neural-networks/sphere.html?cube

Install script

Download the code using GIT

git clone https://github.com/wasalm/2-valued-neural-networks
2-valued-neural-networks

Generic setup

With the following lines we setup the virtual environment and install the needed libraries:

python3 -m venv --system-site-packages ./environment
source ./environment/bin/activate
python -m pip install -U pip
python -m pip install numpy wheel
python -m pip install matplotlib

Installation of jax

Depending on machine, we need to install a different version of jax.

MacOS

If we run on MacOS and we want to use the experimental driver, we run

python -m pip install jax-metal

A computer with a graphics card (CUDA)

If we have a graphics card available that supports CUDA, we run

python -m pip install "jax[cuda12]"

Other cases

In any other case we run

python -m pip install jax

Running of code

To run the code, we run

source ./environment/bin/activate
./train-local.sh