pyro-ppl/pyro

extending Predictive to support sampling by plate name

Open

#2,783 opened on Mar 16, 2021

View on GitHub
 (5 comments) (1 reaction) (0 assignees)Python (981 forks)batch import
discussionenhancementhelp wanted

Repository metrics

Stars
 (8,211 stars)
PR merge metrics
 (Avg merge 10d 19h) (1 merged PR in 30d)

Description

Following our earlier discussion with @fritzo and @adamgayoso , creating this issue. Would be great if Predictive class can support sampling of all variables in a given plate, automatically pulling all site values in a given plate(plate_name) or all sites not is that plate, including both samples and observed values:

trace = poutine.trace(model).get_trace(*args, **kwargs)  # you'll need to fill in args,kwargs
cell_samples = {
    name: site["value"]
    for name, site in trace.nodes.items()
    if site["type"] == "sample"
    if any(f.name == plate_name for f in site["cond_indep_stack"])
}

This functionality would simplify dealing with local variables within a minibatch plate, by making it easy to do separate predictive calls for the minibatch plate variables and global variables.

Contributor guide

Research direction
Explore the Predictive class in Pyro's source code to understand how it currently samples. Implement a new parameter 'plate name' that filters sites based on cond indep stack. Modify the forward method to only include sites matching the plate. Test with a simple model that has a plate and verify that only plate sites are sampled.
Tech stack
python
Domain
ai
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
2
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
Half day
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
PythonPyro
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
60