pymc-devs/pymc

Add function that goes from transformed space to untransformed space

Closed

#6,721 opened on May 18, 2023

View on GitHubContributor guide
 (3 comments) (0 reactions) (0 assignees)Python (1,902 forks)batch import
enhancementshelp wantedtrace-backend

Repository metrics

Stars
 (7,926 stars)
PR merge metrics
 (Avg merge 11d) (7 merged PRs in 30d)

Description

Description

Because we don't save transformed variables in the returned InferenceData (why not?) it's not easy to evaluate the model logp once we have a trace.

One could rewrite the model without transforms (and we can make this automatically for the user) This is possible with https://www.pymc.io/projects/docs/en/stable/api/model/generated/pymc.model.transform.conditioning.remove_value_transforms.html

But someone might still want to evaluate it in the original model (with jacobians and all that).

One dirty implementation is given here: https://discourse.pymc.io/t/logp-questions-synthetic-dataset-to-evaluate-modeling/12129/6?u=ricardov94

Contributor guide

Research direction
Investigate the existing `remove value transforms` function in pymc.model.transform.conditioning to understand current support for evaluating without transforms. Review the Discourse post linked in the issue for a potential implementation approach. Examine how transformed variables are stored in InferenceData to determine the best way to expose an inverse transform function.
Tech stack
python
Domain
machine learning
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
PythonPyMCBayesian inferencetransforms
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40