Add to requirements-dev.txt and MvNormalSVD method #429
Conversation
|
I think this will break imports for users who have pytensor<2.28.0, because Edit: No need, tests are failing for exactly this reason. I guess we need a try/except, unless @ricardoV94 can think of something more elegant. |
|
PyMC is not yet compatible with this PyTensor version is it? When it is, you'll have to lower pin pymc-extras to the most recent version of PyMC that's compatible with it. |
|
Thanks @jessegrabowski @ricardoV94 for checking this. Got ahead of myself here by not running compatibility tests (rookie mistake). Will go ahead and close this PR. Just a quick question though, how come we only run test on single versions (via .github/workflow/tests.yml): pymc-extras/.github/workflows/test.yml Line 22 in ec46270 Why not include multiple versions like ["3.10", "3.11", "3.12", "3.13"] in the test to broaden the upper pin? I'm guessing there are risks to doing this but am not completely sure |
|
I think it's just to minimize the amount of CI we use, since the tests can be pretty slow (admittedly this is my fault, the statespace tests suck). In principle I agree it would be good to test on as many versions (and OS) as possible. |
There's always scope for more, why not run on all python and numpy and scipy versions allowed by the requirements? Because it has a cost (even if we're not the ones paying it) |
|
makes sense. thanks @jessegrabowski @ricardoV94 for clarifying! |
closes #428
Let me know if opening a git issue wasn't necessary. Wasn't sure if I simple PR would do.