mavenets

Tools for training neural networks on MAVE datasets and running Markov Chain Monte Carlo simulations.

Installation

Installation is supported through a combination of conda and pip. Here is an example setup that will install a compatible environment into ./env and installs the package using cuda 12.4. See pytorch and pyg websites for more detailed installation options.

conda create --prefix ./env python==3.9 pandas pytorch torchvision torchaudio pytorch-cuda=12.4 torch-scatter einops mdtraj numpy -c pytorch -c nvidia -c conda-forge
conda activate ./env
conda install pyg -c pyg
pip install triton
mkdir -p ./src
git clone git@github.com:alekepd/mavenets.git ./src
pip install -e ./src

Usage

Example scripts are provided in ./src/mavenets/example. They can be invoked in the shell for easy usage. For example:

from mavenets.example import run_mlp
run_mlp.scan()

will launch a sample hyperparameter scan over possible multilayer perceptron architectures.