Fix CI errors in statspace test suite #251
Conversation
- Squeeze result vector-matrix multiplication with (1, 1) matrix to avoid shape error in numpy 1.25.2
… scipy.stats._distn_infrastructure)
|
I think the settings files for pymc-experimental need to be consolidated -- we have a pyproject.toml, pytest.ini, and setup.cfg, and all of them have pytest flags inside. I guess only pytest.ini is actually used, though. |
jessegrabowski
force-pushed
the
fix_ci_failures
branch
from
e8fde08 to
2a5fa08
Compare
|
Down to one error, but it's in postprocessing_backend: Literal["cpu", "gpu"] | None = None,Is not 3.9 compatible (the pipe union operator for type hints was added in 3.10). We can either skip the test here, or patch it on the pymc side. Thoughts @twiecki @ricardoV94 ? |
|
@jessegrabowski Let's xfail it here, merge, and fix it on pymc. |
…c.sampling.jax Skip pathfinder test if python < 3.10 Add some comments to `pyproject.toml` to explain what warnings are being ignored and why
|
Looks like there's already a patch in the works here: pymc-devs/pymc#6944 |
|
Make sure those config files I removed weren't there for some other reason (related to building for instance). They seemed to just have pytest and pre-commit settings that were being ignored anyway |
|
You can test that by running |
|
Worked fine, I'm merging this. |
The statsmodels import in the statespace tests are causing everyone's CI to break. This PR fixes that.