Sylwester Arabas comments

Results 518 comments of


                                            Sylwester Arabas

refactor Numba backend to use `@cached_property` instead of defining private methods within ctors (+ phase out `conf.JIT_FLAGS` in favour of ctor args)

this was largely addressed in #1299 an indicator of when the refactor is done would be the removal of `JIT_FLAGS` global constant as of now, there are still some remaining...

add smoke test checking error measures for the Shima_et_al_2009 box example. Closes #327

TODO: check exception type via mark.xfail

Implement merge shuffle algorithm for Numba

Thank you, @zengraf! CC: @piotrbartman I suggest to: - make it an option to use MargeShuffle (`Collision` dynamic `__init__` method seems a good place to set it) - raise a...

Merge PySDM-examples back into PySDM

- [x] re-create examples PRs as PySDM PRs - [x] fix links to documentation to include PySDM or PySDM_examples prefix - [x] old repo: add PySDM and PySDM-examples version specification...

Implement merge shuffle algorithm for Numba

While looking at the changes today with @abulenok (trying to understand the reasons for the breakup test failures), we came up with the idea to create a unit test aimed...

Merge PySDM-examples back into PySDM

all what was reasonably needed to be done is done here, closing

Implement merge shuffle algorithm for Numba

@piotrbartman Thanks for following up! @abulenok Thanks for spotting it and clarifying the issue here! @zengraf Thanks for the new implementation - without it, the old bug would not be...

refactor moments so that histograms are done with one invocation (and check for spurious allocations in products)

As of now, ``moments_body`` takes almost 30% of cpu time in Yang example based smoke tests (e.g., ``test_just_do_it``) with call counts of hundreds of thousands... while in principle all what...

refactor moments so that histograms are done with one invocation (and check for spurious allocations in products)

keep open, this is a few-LoC/high-impact candidate!

refactor moments so that histograms are done with one invocation (and check for spurious allocations in products)

we've just profile a few 0D, 1D and 2D examples with @mstach60161 and this seem not to be an issue anymore! (likely thanks to moments refactors and added multi-threading done...