This repository contains some boilerplate code for running simple machine learning simulations that are intended to be reproduced for some purpose. Functionality is provided for easily decoupling the code concerning data, models and training algorithms, tracking custom made metrics throughout the training process, and finally, running "embarassingly parallel" simulations in a reproducible manner on the specified cpus or gpus. The code is built on top of ignite, which is still actively developed; hence, some of the functionality introduced in this repository might be redundant (or will become so in the future).
I share this code between my projects (e.g., see early-stopped-mirror-descent). Backward compatibility is not intended to be maintained. Anyone that finds the code contained in this repository of any use is welcome to use it without crediting me.