pyro-ppl / pyro-ppl/numpyro

Allow setting initial values on autoguides with subsampling

Open
#2,063 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

enhancement
Dominant language
Python
Stars
2.8k
Forks
316
Avg merge
3d 9h
Merged PRs (30d)
27

Description

Feature Summary

With a model that allows subsampling (subsample_size=n) it is currently not possible to set initial values on autoguides on parameters that have the same dimension as all data with size N, such that n is a subsample of N.

Why is this needed?

This may be useful for complicated (time series) models where it is useful to set initial values for initializing SVI and one wants to use subsampling. (And one does not use amortization)

See here for a MRE: https://forum.pyro.ai/t/svi-autoguides-subsampling-and-initial-values/

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Use the linked forum MRE as the first reproduction case, then trace NumPyro's autoguide handling of initial values when a model uses subsampling. The work is done when the MRE can set initial values for parameters whose full-data dimension exceeds the subsample size and SVI initializes successfully.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
38/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.