stan-dev / stan-dev/loo

Averaging predictive distributions

Open
#83 9 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

feature
Dominant language
R
Stars
157
Forks
38
Avg merge
4d 16h
Merged PRs (30d)
2

Description

This is related to #82 but should be it’s own issue.

Currently we rely on users to use the model weights to create the appropriate mixture of predictive distributions. We should recommend (and demonstrate in a vignette with rstan and loo, and automate in rstanarm and brms) a method for doing this.

There are various options for doing it.
Today @avehtari and I discussed this and we are leaning towards the approach of taking (approx) S*weight_k draws from each posterior predictive distribution, where there are K models and S is the desired sample size. This is better when weights are small than sampling from each posterior predictive distribution with probability weight_k, and easier to implement than a stratified version (maybe an option at some point?).

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

Start by reviewing related issue #82 and the proposed allocation of approximately S*weight_k draws from each posterior predictive distribution. Then examine the loo vignette and the rstanarm and brms integration points; done means recommending and demonstrating a method with rstan and loo and automating it in both integrations.

Written by the indexing model from the issue text.

Assessment

Tech stack
r
Domain
machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.