pymc-devs / pymc-devs/ptgp

Live training monitor dashboard

Open
#7 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Jupyter Notebook
Stars
8
Forks
4
Avg merge
2d 2h
Merged PRs (30d)
4

Description

A TensorBoard-inspired training dashboard for ptgp that renders live charts as a model trains. Scoped to GP training, where the set of useful things to watch is small and known in advance.

This is one of the motivations for using namedtuples as optimization outputs. Each field becomes a plottable series automatically. The existing loss namedtuples (MLLTerms, ELBOTerms, CollapsedELBOTerms) give per-component loss decomposition for free.

Beyond the loss components, some ideas for what else to track:

  • Hyperparameter trajectories (ls, eta, sigma)
  • Variational parameter summaries for SVGP
  • Inducing point movement, or a live scatter plot of Z positions where the user can select which two input dimensions to view
  • Gradient norms per parameter group

Open questions

  • Should this start as a single-user, single-model dashboard? Or target multi-experiment comparison from the start?
  • What else would be useful to plot live during GP training?

Contributor guide

No contributing guide indexed for this repository

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, tests, or entry points are identified. Start by reviewing the existing loss namedtuples (MLLTerms, ELBOTerms, and CollapsedELBOTerms) and the GP training flow. Define the dashboard scope and live series before implementation; done means a working live monitor for the agreed single- or multi-experiment scope.

Written by the indexing model from the issue text.

Assessment

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

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.