SwarmSplitLearning

In this repository, we introduce our own fully distributed variant of the well-known split learning algorithm.

Installation

For a faster installation, install the libmamba-solver first before creating the conda environment:

conda install -n base conda-libmamba-solver
conda config --set solver libmamba

Otherwise, the default conda solver will be used, which may take forever.

Update the file path to your desired directory for the environment by changing the value of prefix: at the end of the environment.yml file. Then run:

conda env create -f environment.yml
conda activate [ENV_NAME]

For tracking experiments using Weights and Biases, place your API key without any spaces in the wandb_key.txt file and make sure that wandb: True is set in the wandb config. Otherwise, metrics etc. will be printed to the console.

If you plan to commit to this repository, please install pre-commit for a consistent code formatting upon committing. Therefore, run the following command in the repo:

 pre-commit install

Optionally, for formatting without committing, you may run:

pre-commit run --all-files