probml / probml/dynamax

Feature request: per-timestep observation uncertainty for Gaussian HMM / ARHMM (heteroscedastic emissions)

Open
#446 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
1k
Forks
114
Avg merge
19h 14m
Merged PRs (30d)
1

Description

Hi there,

I’m fitting HMM/ARHMM models to latent trajectories from an upstream encoder that outputs:

  • mean vector per timestep
  • uncertainty (covariance/variance) per timestep

I’m looking for support for known, time-varying observation uncertainty in emissions, e.g.:

y_t given z_t = k is Normal(mean_{t,k}, Sigma_k + R_t)

where:

  • Sigma_k is state-specific emission covariance
  • R_t is known per-timestep uncertainty from the encoder

Questions:

  1. Does this already exist in Dynamax?
  2. If not, is there a recommended extension point (custom emissions / inputs)?
  3. Is this feature on the roadmap?

If helpful, I’m happy to prototype a PR with guidance on preferred API shape.

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

No files or tests are named. Start by locating the Gaussian HMM and ARHMM emission entry points and any custom-emission or input extension points, then assess how known per-timestep uncertainty could support the stated Normal(mean, Sigma_k + R_t) behavior. Done would be a reviewed API and implementation supporting time-varying observation uncertainty.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.