Population Based Training

I’m interested in using population-based training as a WandB Sweep methodology as referenced in the article here.

However, I can’t find any mention or tutorials for PBT with WandB outside of this article. I’m interested in using PBT with a self-supervised learner and configuring the sweep in a YAML. Does anyone have any advice on how to go about this? I’m familiar with configuring sweeps in a YAML or jupyter notebook but, I’m unsure how to make the PBT scheduler and tuner (as outlined in the article) compatible with the SSL code I’m using which is run from the command line.

Hi Anmol!

Thanks for reaching out. We don’t really have tutorials on PBT, but I found this repo with an implementation on PBT : GitHub - voiler/PopulationBasedTraining: A simple PyTorch implementation of Population Based Training of Neural Networks., which should give you something to start with to integrate into your code.

Is there anything specific to W&B you are strugging with which I can help out with here?

Thanks,
Ramit