Reenable tests that were mostly good already #4830
Conversation
|
The test_distributions.py will be significantly faster after #4736 I'll try to split the speedup part into a separate PR |
michaelosthege
force-pushed
the
reenable-tests
branch
from
d820774 to
028ca22
Compare
michaelosthege
force-pushed
the
reenable-tests
branch
from
2dd47fd to
a28a582
Compare
pymc3/tests/test_profile.py
Outdated
| @@ -23,7 +23,7 @@ def test_profile_model(self): | |||
| assert self.model.profile(self.model.logpt).fct_call_time > 0 | |||
|
|
|||
| def test_profile_variable(self): | |||
| assert self.model.profile(self.model.value_vars[0].logpt).fct_call_time > 0 | |||
| assert self.model.profile(self.model.value_vars[0]).fct_call_time > 0 | |||
There was a problem hiding this comment.
This one does not seem equivalent...
| assert self.model.profile(self.model.value_vars[0]).fct_call_time > 0 | |
| rv = self.model.basic_RVs[0] | |
| self.model.profile(pm.logpt(rv, self.model.rvs_to_values[rv])).fct_call_time |
There was a problem hiding this comment.
It returns the TensorVariable which is then profiled, no?
Can you suggest a change? As you can see I'm still confused by these logpt things..
There was a problem hiding this comment.
I added a suggestion in my first comment, but it has to be tested locally, as I might have screwed up the names while copy pasting from the REPL
A few XFAILs were added because DensityDist is not yet refactored and a shape issues causes the CompoundStep to fail.
This way they don't XPASS sometimes. Also it doesn't waste compute resources on tests whoose outcome has no relevance.
michaelosthege
force-pushed
the
reenable-tests
branch
from
a28a582 to
d0ec6fb
Compare
|
This failing test seems pretty close to the expected result https://github.com/pymc-devs/pymc3/pull/4830/checks?check_run_id=2971381082#step:7:199 But if you keep finding issues it might be related to #4771 |
This PR includes more test files in the CI actions.
Only three remain excluded:
test_types.pywhere I wasn't familiar enough with what's going on internally (see Reenable test_types.py and make it run faster #4828)test_mixture.pytest_variational_inference.pyI also split up the test jobs based on their runtime. This was before, labeled by the name of the first test in each job.
The
test_distributions.pyis by far the largest single one. We should definitely split it into two or three files and run them separately!!