Issue page for xfails added during aeppl integration.

[kc611]

The XFails added during aeppl integration #4887 :

• pymc/tests/test_idata_conversion.py::TestDataPyMC::test_multivariate_observations : Already fixed. Xfail decorator can be removed on next release of Aesara.
• pymc/tests/test_transforms.py::TestElementWiseLogp::test_beta_ordered : Already fixed. Xfail decorator can be removed on next release of Aesara.
• pymc/tests/test_sampling.py::TestSamplePPC::test_deterministic_of_observed: Failing on specific systems similar to those above. i.e. Float32 Linux systems. This was caused by the new Assertions introduced by aeppl and discussed in AssertionError in aeppl.logp breaks sampling in pymc aesara-devs/aeppl#84. Unpin aeppl version #5203 Fixes this by making the prior more lenient.
• pymc/tests/test_posteriors.py::TestSliceUniform : Failing on specific systems similar to those above. i.e. Float32 Linux systems. Fixed in Unpin aeppl version #5203
• pymc/tests/test_transforms.py::test_interval_near_boundary : This test produces infinite values using aeppl. Fixed in Unpin aeppl version #5203
• pymc/tests/test_transforms.py::test_chain_jacob_det : Fails due to precision issue. Values just close to expected. (Deviation of order 10^-6).
  • The log_jac_det of the LogOdds transform is slightly different than the previous implementation and less comparable to the one given by actual_ljd function. This difference could be due to either a lower or higher numerical precision of the new log_jac_det, and it is unclear which one it is without access to an evaluation with known error bounds.

We can discuss these over here.

[twiecki]

@brandonwillard will cut a new aesara release so that should help with some of these.

[twiecki]

@kc611 we have a new aesara release: https://github.com/aesara-devs/aesara/releases/tag/rel-2.2.3 (thanks @brandonwillard !)

[ricardoV94]

Closing this given given no complaints/ low priority/ need for investigation