Jan comments

Results 241 comments of

Jan

Computation of denoising posterior precision matrix in JAC method (score_fn_iid)

Thanks for raising this @touronc ! @manuelgloeckler any thoughts?

FMPE fails to sample with 2D conditions

Thanks for raising this @gmoss13 ! Your proposed solution sounds good. 🚀

Logit transform

@anastasiakrouglova thanks for the updates! 🙏 is this ready for another round of reviews?

Added Apache License reference comments to source files + CI bash script check

> Cool! We could also have a github workflow that checks this, as, for example, [here](https://github.com/jaxleyverse/jaxley/blob/3ea78e34329abaf0f5d53d70808124b22e0e0313/.github/workflows/tests.yml#L43-L81). @nMaax would you be up for adding something like this to the `sbi` CI...

FMPE not working even on simple tasks?

Thanks for reporting this! Indeed, for the mini-sbibm we are working on, see #1335 , we observe similar performance on `two-moons` (with fewer simulations though). We will have a look...

FMPE not working even on simple tasks?

there was a potentially relevant bug fix: #1492

FMPE not working even on simple tasks?

See this PR #1497 for potential improvements. e.g., this comment about the architecture: https://github.com/sbi-dev/sbi/pull/1497#issuecomment-2751932097

FMPE not working even on simple tasks?

#1544 add bug fixes and improvements to these methods and fixed many performance issues.

`simulate_for_sbi` not saturating CPU

Dear @janko-petkovic Thanks a lot for looking into this and proving the benchmark. I can confirm your findings. This is indeed a very interesting behavior of `simulate_for_sbi`. The "embarrassingly parallel"...

`simulate_for_sbi` not saturating CPU

Looking into the code for `simulate_for_sbi` and `simulate_in_batches`, I also see that it really needs some refactoring. I will make a PR. If you like you can contribute the `joblib`...