Over the past two days I went ahead and made a basic wrapper for orchestrating research experiments. I named it skeletor. It is a simple add-on to standard ML training code. 99% of the work is being done by two other python modules, track and ray. This just combines them to let you easily log experiments, launch a bunch of experiments in parallel, and then analyze them all together in a nice pandas DataFrame. I also tried to add in some basic model, dataset, and optimizer loading code to reduce boilerplate a bit, and I’ll add more later. You can install it with
pip install skeletor-ml.