Closed rsokl closed 2 years ago
Hi there! Thank you for bringing this to my attention, hydra-zen looks interesting. I'll definitely keep this in mind for investigation as part of a future release.
Would definitely require a large re-write. I am not entirely sure if hydra-zen would support some of the features I need with regards to overriding and downstream projects. Will definitely need to investigate this too.
I am not entirely sure if hydra-zen would support some of the features I need with regards to overriding and downstream projects. Will definitely need to investigate this too.
I'm happy to provide insights towards this end if you have any questions.
Hello! I just came across your project - it looks really nice!
I noticed that you are using Hydra's yamls to make your experiments configurable. I figured I might put hydra-zen on your radar.
hydra-zen provides simple functions for dynamically creating dataclass-based configs (a.k.a structured configs), and it is especially useful for large-scale projects like yours.
There are a few big "wins" to this:
disent
).For example, this is all you need to generate configs for encoder and decoder models:
You could then use simple utility functions to inject globally-specified settings (e.g.
x_shape='${dataset.meta.x_shape}'
) into instances of these configs. This really helps keep you from having to repeat yourself.I hope this is somewhat helpful (or at least interesting). Just wanted to bring this to your attention - feel free to close this issue 😄