goktug97 / de-torch

Minimal PyTorch Library for Differential Evolution

Minimal PyTorch Library for Differential Evolution

Requirements

numpy
torch

Optional

mpi4py # If you want to run in parallel
gym # For examples

To install mpi4py you need MPI installed in the system. Check: https://mpi4py.readthedocs.io/en/stable/install.html

A Dockerfile is provided for convenience.

Installation

pip install detorch --user

Source

git clone --recursive https://github.com/goktug97/de-torch
cd de-torch
python3 setup.py install --user

Usage

See https://github.com/goktug97/de-torch/blob/master/examples

Check https://github.com/goktug97/de-torch/blob/master/detorch/config.py for parameters.

You can run the example with

PYTHONPATH="$(pwd):$PYTHONPATH" python examples/example.py

or in parallel for faster training (requires mpi4py).

PYTHONPATH="$(pwd):$PYTHONPATH" mpirun -np 2 python examples/example.py

Another Evolution Library for PyTorch

https://github.com/goktug97/nes-torch

de-torch's Issues

How to mutate images

Thank you. I am currently checking out the possibility of using DE in my thesis. This paper is using DE to perturb a pixel. The DE implementation is from Scipy. How is yours different from theirs? can I use the one in this repo like it's used in the paper? Thanks