Michael Deistler comments

Results 46 comments of


Michael Deistler

log_prob fails for Restricted Prior when sampling with SIR

Got it, thanks! The quick solution for now is to **not** run `process_prior` on the `RestrictedPrior`. If you follow the tutorial [here](https://www.mackelab.org/sbi/tutorial/16_implemented_methods/), everything should work: ```python from sbi.inference import SNPE...

log_prob fails for Restricted Prior when sampling with SIR

Thanks for clarifying! I agree that we should allow passing the RestrictedPrior to be passed at unit. I will fix this soon, but I don’t have time this week. Until...

log_prob fails for Restricted Prior when sampling with SIR

Yes, that's only true for SNPE.

log_prob fails for Restricted Prior when sampling with SIR

For now, I recommend simply passing the original prior at initialization. I shouldn't make a big difference. I'll fix this next week.

Use conventional commits and commit linter?

Depends on how well it works I guess :) If it just copies PR descriptions (or commit titles) then I think the resulting CHANGELOG will not really helpful, but if...

Preconfigured embedding_nets

We have CNNs, MLPs, and IID-nets, so I am closing this for now.