search

LSSTDESC / parsnip

Deep generative modeling of astronomical transient light curves
MIT License
7 stars 10 forks source link

Reproducibility #5

Open hombit opened 2 years ago

hombit commented 2 years ago

What do you think about an idea of adding an option of parsnip_train to set random seeds to make training reproducible?

kboone commented 2 years ago

That's a great idea.

We use random sources from both numpy (for augmentation) and pytorch (for initializing the model and for sampling from the VAE). Presumably we could set the seeds for both of those in parsnip_train and that would make things reproducible.