SciML / SciML/Optimization.jl

Let OptimizationPolyalgorithms return or save the optimization state

Open
#307 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Julia
Stars
839
Forks
100
Avg merge
20h 43m
Merged PRs (30d)
38

Description

Currently, the final state of the Adam and BFGS is dropped when solve stops. However, it can be beneficial to resume training, and then it would help to resume Adam and BFGS from their last settings.
This could be handled by making PolyOpt an object, as when using Adam from Flux.

As an example:
opt = PolyOpt()
// Start training
solve(problem, opt);
// Change something - e.g. adding data points when fitting a neural ode
problem = ...
// Softly restart training
solve(problem, opt);

opt then stores the relevant information.

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 tracing OptimizationPolyalgorithms and the solve entry point, focusing on how Adam and BFGS state is currently created and discarded. Compare the proposed PolyOpt object with Flux's Adam state handling, then define how repeated solve calls preserve state and verify that training can resume after the problem changes.

Written by the indexing model from the issue text.

Assessment

Tech stack
julia
Domain
api, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.