Support PyMC v5 #309

vandalt · 2023-11-03T05:53:31Z

This builds upon #271, where pretty much all the hard work required to update from PyMC3 had already been done, and implements support of PyMC v5. Most of the changes were replacing aesara by pytensor plus a few API changes.

Some things left to address:

Should the joss directory be updated? I'm guessing the answer is no
Update requirement version for exoplanet-core, pymc-ext and celerite2 once pymc >= 5 support is released there. This is required for the tests to pass.
Is it OK to remove warning filters in utils.docs_setup()? I left TODO comments to discuss this.
There is a reference to pytensor.opt.Assert (which used to be theano.opt.Assert) which no longer exists. See comment in the code
For the Docs: is the lowest compatible Python version 3.6? I have not tested this but I saw that 3.8 is the lowest version test by nox (and the lowest still getting security updates I think)
Updating the case studies. Not technically required before merging into main, but probably worth waiting in case it exposes bugs or deprecation warnings.

src/exoplanet/light_curves/limb_dark.py

src/exoplanet/utils.py

docs/user/install.rst

dfm · 2023-11-07T12:12:30Z

@vandalt — Thanks!! 🚀

It will take me a week or so to go through this in detail, but here are some initial comments.

First, can you take a look at the merge conflicts with main? You could rebase if you want, but I'd also be fine with the merge commit (merge from main, fix the conflicts, then commit).

Should the joss directory be updated? I'm guessing the answer is no

Agreed. I think we should leave that as is.

Update requirement version for exoplanet-core, pymc-ext and celerite2 once pymc >= 5 support is released there. This is required for the tests to pass.

I'll try to get those released ASAP. Might be worth adding a test against the GitHub versions of those in the meantime as well though? What do you think?

Is it OK to remove warning filters in utils.docs_setup()? I left TODO comments to discuss this.

This is fine with me as long as pytensor doesn't spew so many warnings!

For the Docs: is the lowest compatible Python version 3.6? I have not tested this but I saw that 3.8 is the lowest version test by nox (and the lowest still getting security updates I think)

I'm happy to update to 3.8 or 3.9 as the lower bound!

Updating the case studies. Not technically required before merging into main, but probably worth waiting in case it exposes bugs or deprecation warnings.

I think this is a big one, but I don't think we should worry about solving it before merging this PR. Instead, we should open a separate issue to track that.

vandalt · 2023-11-07T16:15:16Z

Thanks!

I'll try to get those released ASAP. Might be worth adding a test against the GitHub versions of those in the meantime as well though? What do you think?

Yes, I can add that!

This is fine with me as long as pytensor doesn't spew so many warnings!

I'll try removing and make sure there are not too many warnings when running the tests with pytest -s or the tutorials

I think this [case studies update] is a big one, but I don't think we should worry about solving it before merging this PR. Instead, we should open a separate issue to track that.

I agree it's a big one, but I think trying a quick update+run for each notebooks (without necessarily fixing everything) would be worthwhile. So far the pRV one helped uncover a problem with passing observed=ecc to eccentricity priors (basically having observed pointing to a prior or Deterministic parameter no longer works. We need to use pm.Potential). I added a test for this and will push it soon, hopefully with a fix

vandalt · 2023-11-07T21:52:41Z

@dfm another point I forgot to raise is that unit_disk and angles are defined both here and in exoplanet-dev/pymc-ext. Should one of them be removed (with a deprecation warning at first)?

vandalt · 2024-03-06T13:48:39Z

Hi @dfm!

I finally took time to look at the remaining issues with this PR.

I just finished a first attempt at updating the case studies. I'll send a PR on that repo later today.
While doing that, I noticed (as mentioned above) that PyMC 5 no longer supports observed to depend on other random variables/model parameters. This required an update of the eccentricity prior to use pm.Potential() when ecc is a derived parameter.
I rebased onto main (not super experienced with rebase/merged conflict, hopefully I did not mess the commit history too much!)

I think the only remaining thing would be to update the celerite2, pymc-ext and exoplanet-core requirements. Let me know if it's best to wait for release candidates or if I should test against the git versions for now.

Thanks!

dfm · 2024-03-06T14:52:58Z

Thanks for all your work on this, @vandalt!!

I'll work on getting those other packages released ASAP, but it probably won't be until next week. Can you try to remember to ping me if you don't see any motion by Monday afternoon or Tuesday morning next week?

dfm · 2024-03-07T14:29:00Z

@vandalt — I did manage to publish new releases of all of the dependencies yesterday and re-ran the tests. Any guesses for why the PyMC3 tests are failing now?

vandalt · 2024-03-08T22:43:32Z

@dfm

Yes sorry, I had forgotten to re-run the tests with PyMC3 locally. Most errors were related to differences in how to access the shape of a variable in PyMC3 vs 5.

For the Kipping eccentricity prior, I had added a test for something that was not implemented in the PyMC3 version (truncated distribution with observed eccentricity), so I modified the test to expect the NotImplementedError.

For the vaneylen19 eccentricity prior, I'm not 100% sure what caused the issue, but reverting the PyMC3 version of the prior closer to its original form fixed the issue on my local setup. This comes at the cost of slightly more if USING_PYMC3 conditions, but at least now the tests pass for both versions!

dfm · 2024-03-08T23:25:31Z

Great - thanks!! I'll look into debugging the docs build errors because those can be tricky to track down.

vandalt · 2024-03-09T03:30:26Z

It seems the docs thing was not too tricky (and 100% my fault 😅): I had moved exoplanet-core from the "main" dependencies to the pymc and pymc3 extra groups, because the required versions were now different. I just added exoplanet-core to the docs extra as well.

for more information, see https://pre-commit.ci

In PyMC 3, we could just used `observed=ecc` with any distribution. In PyMC >= 5, we need to use a `pm.Potential`

Same as kipping13: if eccentricity is "observed" (derived), extra prior must be a `pm.Potential`

Previously, default initval was hardcoded so passing shape without initval caused error. Added failing tests and fixed

…nds in kipping13 This did not work with the original PyMC3 implementation either, but is supported in PyMC 5. This error makes it a bit clearer

…stribution

for more information, see https://pre-commit.ci

`type.shape` works only in PyMC > 3

The implementation that worked with PyMC v5 did not in PyMC3

Had to remove from the main dependencies because no of the different versions for PyMC3 and PyMC 5

dfm · 2024-04-16T15:25:20Z

@vandalt — I pushed a few changes to fix some merge conflicts and get the tests passing, but now I think we're good to go here!! I'm going to merge and we can fix issues as they come up in future PRs. Thanks again for all your work getting this over the finish line!! I (and the whole community) really really appreciate everything that you've done here!!! 🚀

vandalt mentioned this pull request Nov 3, 2023

Use model argument in Evaluator when getting model context exoplanet-dev/pymc-ext#39

Merged

vandalt commented Nov 3, 2023

View reviewed changes

src/exoplanet/light_curves/limb_dark.py Outdated Show resolved Hide resolved

vandalt commented Nov 3, 2023

View reviewed changes

src/exoplanet/utils.py Outdated Show resolved Hide resolved

vandalt commented Nov 3, 2023

View reviewed changes

docs/user/install.rst Outdated Show resolved Hide resolved

dfm mentioned this pull request Nov 28, 2023

Conflicts with current version of arviz/xarray #310

Closed

vandalt force-pushed the pymc5 branch from 05b9b41 to 612a74c Compare March 6, 2024 13:34

vandalt mentioned this pull request Mar 6, 2024

Update case studies to pymc v5 exoplanet-dev/case-studies#78

Open

dfm and others added 14 commits April 16, 2024 10:36

Implementing support for pymc v4

d2a77df

[pre-commit.ci] auto fixes from pre-commit.com hooks

6e93721

for more information, see https://pre-commit.ci

workflow error

8fb3619

workflow

7b4d1ee

allow python version for lint

4faf203

getting distributions to work

33402fe

some tweaks for pymc3

c2db519

[pre-commit.ci] auto fixes from pre-commit.com hooks

5b243d9

for more information, see https://pre-commit.ci

pymc3 fixes

a42f7a0

[pre-commit.ci] auto fixes from pre-commit.com hooks

fbeb4b8

for more information, see https://pre-commit.ci

refactoring distributions again

9707898

docstrings for distributions

6bba73c

starting to re-write docs

72c337a

[pre-commit.ci] auto fixes from pre-commit.com hooks

8d85a1b

for more information, see https://pre-commit.ci

vandalt and others added 25 commits April 16, 2024 10:36

Update docs from PyMC 4 to PyMC 5 (mostly pytensor -> Aesara)

d9d189b

Comment-out warnings in docs_setup() with TODO to maybe remove them

7cb46ca

TODO comment for deprecated pt.opt.Assert

0824dc9

Update Evaluator() call to use model kwarg

f2929cd

Replace testval by initval and start by initvals

961546c

Enable using kipping13 prior when eccentricity is a derived parameter

6c2c522

In PyMC 3, we could just used `observed=ecc` with any distribution. In PyMC >= 5, we need to use a `pm.Potential`

Add _truncate_dist helper function for eccentricity prior

3447745

Fix vanelyen19 prior when eccentricity is a derived parameter

91d94e4

Same as kipping13: if eccentricity is "observed" (derived), extra prior must be a `pm.Potential`

Test angle() and unit_disk() distributions and fix shape error

5d3b85c

Previously, default initval was hardcoded so passing shape without initval caused error. Added failing tests and fixed

Pop initval when ecc is None only when bounds are set

488051c

Clean-up _truncate_dist() helper function

24d4816

Raise NotImplementedError when trying to pass observed ecc with bou…

ced141e

…nds in kipping13 This did not work with the original PyMC3 implementation either, but is supported in PyMC 5. This error makes it a bit clearer

Test eccentricity priors with and without bounds

85bcce8

Update required Python version from 3.6 to 3.8 in docs

febddcb

Convert shape from int to tuple for pymc>3 in impact parameter di…

fd97806

…stribution

Remove deprecated pt.opt.Assert in favor of Assert from compat

32e6283

Remove leftover commented-out warning filters

054c5bf

[pre-commit.ci] auto fixes from pre-commit.com hooks

38e4938

for more information, see https://pre-commit.ci

Replace tox dependency by nox

37661a0

Check that kipping13_observed raises error with bounds when using PyMC3

6ae26ca

Get shape using tag.test_value for PyMC3 tests

7d37c7f

`type.shape` works only in PyMC > 3

Remove extra potential in vaneylen19 distribution PyMC 3 version

892c307

The implementation that worked with PyMC v5 did not in PyMC3

Update dependency versions for PyMC v5

6284530

Add exoplanet-core dependency to docs

0ef8489

Had to remove from the main dependencies because no of the different versions for PyMC3 and PyMC 5

Another merge conflict

d4c3345

dfm force-pushed the pymc5 branch from 790fad7 to d4c3345 Compare April 16, 2024 14:37

dfm added 2 commits April 16, 2024 10:40

Range of tests for deps

2d0e144

Don't install coverage

0b184e7

dfm merged commit c4eceb4 into exoplanet-dev:main Apr 16, 2024
9 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support PyMC v5 #309

Support PyMC v5 #309

vandalt commented Nov 3, 2023 •

edited

Loading

dfm commented Nov 7, 2023

vandalt commented Nov 7, 2023

vandalt commented Nov 7, 2023 •

edited

Loading

vandalt commented Mar 6, 2024

dfm commented Mar 6, 2024

dfm commented Mar 7, 2024

vandalt commented Mar 8, 2024

dfm commented Mar 8, 2024

vandalt commented Mar 9, 2024

dfm commented Apr 16, 2024

Support PyMC v5 #309

Support PyMC v5 #309

Conversation

vandalt commented Nov 3, 2023 • edited Loading

dfm commented Nov 7, 2023

vandalt commented Nov 7, 2023

vandalt commented Nov 7, 2023 • edited Loading

vandalt commented Mar 6, 2024

dfm commented Mar 6, 2024

dfm commented Mar 7, 2024

vandalt commented Mar 8, 2024

dfm commented Mar 8, 2024

vandalt commented Mar 9, 2024

dfm commented Apr 16, 2024

vandalt commented Nov 3, 2023 •

edited

Loading

vandalt commented Nov 7, 2023 •

edited

Loading