-
-
Notifications
You must be signed in to change notification settings - Fork 2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Bump PyTensor dependency #6830
Bump PyTensor dependency #6830
Conversation
Codecov Report
Additional details and impacted files@@ Coverage Diff @@
## main #6830 +/- ##
==========================================
- Coverage 92.05% 92.03% -0.02%
==========================================
Files 96 96
Lines 16373 16378 +5
==========================================
+ Hits 15072 15074 +2
- Misses 1301 1304 +3
|
2e57d53
to
69927c8
Compare
Had to pin arviz from above, as it was installing on the python 3.8 and failing |
68fa09a
to
0783c4d
Compare
0783c4d
to
3cc3522
Compare
Let's just drop 3.9 too. |
3.9 is supported by Numpy until next January. I think Oriol meant 3.8 in the comment he left, if that's what led you to say that? |
I did mean 3.8, sorry |
Last PyTensor release broke something on VI @ferrine. https://github.com/pymc-devs/pymc/actions/runs/5585372543/jobs/10208039570?pr=6830 My guess it has to do with this one: But could also be an interaction with this one: Looking at the Any chance you can take a look? |
f8cfe35
to
d97d587
Compare
d97d587
to
d518259
Compare
d518259
to
e2ec95a
Compare
e2ec95a
to
845c8bb
Compare
Took 3 releases, but CI is passing! :) |
I'm good on the vi side, anything else? |
Thanks @ferrine! Also with the help debugging! I requested another review for the non-VI changes. Will wait a bit for feedback before merging. |
major as in PyMC v6 ? I'm just on the phone until Sunday, so I can't really check the diff. (Don't wait for mel However, I'd like to request an example of a line of code that will fail and how to fix it to be included in the release notes. |
No, major as in PyMC v5.X.0 We are not doing "major, major" releases unless the PyMC API / dependencies change drastically.
We are reintroducing static broadcasting-only again with the PyTensor bump, so for example broadcasting with All other needed changes were internally-only and shouldn't affect users. Mostly, I imagine it will break custom defined MultivariateRVs that relied on the old automatic shape inference.. |
Bumps PyTensor dependency to 2.14.1
This PyTensor release reverts dynamic broadcasting in Elemwise. Code that started to rely on it will fail, hence why it is marked as a major PyMC release as well.
We also keep shape dependencies in the original graph, so some logic that expected eager Constants had to be tweaked.
Finally, no more automatic support shape inference for Multivariate RVs. The thing we had going on wasn't robust and made errors cryptic in the cases where it didn't apply. This includes dropping support for Multivariate CustomDist not implemented with via
dist
📚 Documentation preview 📚: https://pymc--6830.org.readthedocs.build/en/6830/