What is the status of this project?

[turian]

Just curious. It seems cool

[mauvilsa]

The goals that I want to achieve are clear and have been commenting this with people and working on it from time to time. For example there is a recent discussion in the pytorch-lightning's lightning-cli slack channel. Might be that this gets contributed to https://github.com/Lightning-AI/lightning-hpo, but this requires a significant effort which I haven't had the time for.

[turian]

@mauvilsa Ah cool, you're also in Berlin. I'm on their slack, you can ping me there too.

[EthanMarx]

Hey! Was building something similar (see themain branch https://github.com/ethanmarx/lightray) and came across this repo.

I ended up refactoring what I had based on this setup (see the refactor branch https://github.com/EthanMarx/lightray/tree/refactor), since it allowed cleaner configuration of the tuning parameters (e.g. the search algorithm) from the config. One difference is I use the tune.Tuner API as opposed to tune.run. I also added some utilities to better support s3 storage, which is specific to my use cases.

Anyways, wanted to bring this to your attention to get some feedback, and credit you somehow for stealing this. Also curious if someone else has been working on a more robust version of this so I don't have to :)

[mauvilsa]

I still hadn't had time to work on this further. And I don't know if there is any better version of it. Great if you are able to use what I had here. I will look at it when I have some time. It would be great if it automatically saved the tune config similar to what LightningCLI does for a single fit.