NotificationsYou must be signed in to change notification settings
Fork2.2k
Star9.4k

Support PyTensor deterministic operations as observations#7656

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 ourterms of service andprivacy statement. We’ll occasionally send you account related emails.

Already on GitHub?Sign in to your account

Jump to bottom

Merged

ricardoV94 merged 14 commits intopymc-devs:mainfromwilliambdean:data-as-observed

Feb 27, 2025

Merged

Support PyTensor deterministic operations as observations#7656

ricardoV94 merged 14 commits intopymc-devs:mainfromwilliambdean:data-as-observed

Feb 27, 2025

Conversation

Copy link

Contributor

williambdean commentedJan 24, 2025•
edited
Loading

Description

Will need some help on this implementation and how to best test this.

Related Issue

ClosesBUG: data as observed in RV #7649
Related to #

Checklist

Checked thatthe pre-commit linting/style checks pass
Included tests that prove the fix is effective or that the new feature works
Added necessary documentation (docstrings and/or example notebooks)
If you are a pro: each commit corresponds to arelevant logical change

Type of change

New feature / enhancement
Bug fix
Documentation
Maintenance
Other (please specify):

📚 Documentation preview 📚:https://pymc--7656.org.readthedocs.build/en/7656/

add a test case

e3f66d9

github-actionsbot added the bug label

Jan 24, 2025

williambdean added2 commits

January 24, 2025 21:50

check for random ancestors as well

7b39d7a

check for inputs first

b36e573

ricardoV94 reviewed

Jan 25, 2025

View reviewed changes

pymc/data.py OutdatedShow resolvedHide resolved

ricardoV94 reviewed

Jan 25, 2025

View reviewed changes

pymc/data.py OutdatedShow resolvedHide resolved

Copy link

Member

ricardoV94 commentedJan 25, 2025

We need to tell InferenceData converter how to get the data values to put inconstant_data, andobserved_data. It had some hardcode logic assuming you could only due Casts

ricardoV94 added enhancements and removed bug labels

Jan 25, 2025

Copy link

Member

ricardoV94 commentedJan 25, 2025

I don't consider this a bugfix, it's behavior that was explicitly forbidden for conservative reasons (more like a NotImplementedError)

use existing function

7f71397

Copy link

ContributorAuthor

williambdean commentedJan 25, 2025

We need to tell InferenceData converter how to get the data values to put inconstant_data, andobserved_data. It had some hardcode logic assuming you could only due Casts

Where does this happen in the code?

Copy link

ContributorAuthor

williambdean commentedJan 25, 2025

I think the MiniBatch tests will still fail. Any thoughts on how that should behave?

Copy link

ContributorAuthor

williambdean commentedJan 25, 2025

Also, feel free to choose a better title! I couldn't express it too well

eval for the observed_data group

b182686

williambdean commented

Jan 25, 2025

View reviewed changes

tests/test_data.pyShow resolvedHide resolved

Copy link

Member

ricardoV94 commentedJan 25, 2025•
edited
Loading

We need to tell InferenceData converter how to get the data values to put inconstant_data, andobserved_data. It had some hardcode logic assuming you could only due Casts
Where does this happen in the code?

pymc/pymc/backends/arviz.py

Line 63 infa43eba

obs_data=extract_obs_data(aux_obs)

pymc/pymc/pytensorf.py

Line 152 infa43eba

defextract_obs_data(x:TensorVariable)->np.ndarray:

We should be able to just use constant_fold (also in pytensor) for it

ricardoV94 changed the title~~Support ops to pm.Data in observed variables~~Support pytensor deterministic operations as observations

Jan 25, 2025

ricardoV94 changed the title~~Support pytensor deterministic operations as observations~~Support PyTensor deterministic operations as observations

Jan 25, 2025

Copy link

ContributorAuthor

williambdean commentedJan 25, 2025

We should be able to just use constant_fold (also in pytensor) for it

I have an implementation before this suggestion. I need a little help understanding this. Feel free to put suggestion

Copy link

Member

ricardoV94 commentedJan 26, 2025

We should be able to just use constant_fold (also in pytensor) for it
I have an implementation before this suggestion. I need a little help understanding this. Feel free to put suggestion

What do you mean? My suggestion is to simply call constant_fold on the observed variable

Copy link

ContributorAuthor

williambdean commentedJan 28, 2025•
edited
Loading

I am not able to make this work with eitherconstant_fold(x) orconstant_fold([x]). Both return errors. Is the constant_fold from pymc.pytensorf or from pytensor?

Copy link

Member

ricardoV94 commentedJan 28, 2025

I am not able to make this work with eitherconstant_fold(x) orconstant_fold([x]). Both return errors. Is the constant_fold from pymc.pytensorf or from pytensor?

From pymc.pytensorf. There's araise_if_not_constant flag you can set, but what cases is it failing?

Copy link

Member

ricardoV94 commentedJan 28, 2025

Ah if it's a SharedVariable likepm.Data of course constant_fold is not going to work... Dummy me.

Copy link

Member

ricardoV94 commentedJan 28, 2025•
edited
Loading

frompytensor.compile.modeimportModeimportpymcaspmwithpm.Model(coords={"date": [0,1,2]})asm:data=pm.Data("data", [0,1,2],dims="date")x=pm.Normal("x")y=pm.Normal("y",x,observed=data,dims="date")cheap_eval_mode=Mode(linker="py",optimizer=None)m.rvs_to_values[y].eval(mode=cheap_eval_mode)# array([0., 1., 2.])

specify the mode

b2a3d1e

Copy link

ContributorAuthor

williambdean commentedJan 28, 2025

Thanks for the suggestion. I've used mode on the random variable directly. Is it different to get the rvs_to_values from the model? it currently isn't accessed from the function at the moment.

Copy link

ContributorAuthor

williambdean commentedJan 28, 2025

Could you point me to similar tests for this?

add test to extract function

acd22f3

Copy link

ContributorAuthor

williambdean commentedJan 28, 2025

I've just added to associated pytensorf test suite

williambdean deleted the data-as-observed branch

February 27, 2025 10:50

williambdean mentioned this pull request

Mar 14, 2025

Disconnected node in model graph after deterministic operations#7722

Closed

ricardoV94 mentioned this pull request

Apr 28, 2025

Don't allow Deterministics in the graph of observed#7769

Open

vandalt pushed a commit to vandalt/pymc that referenced this pull request

May 14, 2025