FluxML/Flux.jl

Add docs note about saving/loading models with anonymous functions

Open

#2263 opened on Jun 9, 2023

View on GitHub
 (9 comments) (1 reaction) (0 assignees)Julia (4,725 stars) (619 forks)batch import
documentationhelp wanted

Description

The new save/load docs promote JLD2.jl which does not support saving/loading anonymous functions reliably. This most commonly occurs for activation functions. The solution is to use Flux.state + Flux.loadmodel! and set the desired anonymous function in the destination of loadmodel!. This avoids needing the serialization library to correctly handle the anonymous function.

This will be problematic when the anonymous function contains data (state) that actually should be restored. A possible solution here is to make the closure an explicit struct. Maybe there are better solutions.

Regardless, new users are unlikely to realize these edge cases. We should expand the saving/loading documentation to explain how to handle these cases with code examples.

Contributor guide

Tech stack
None
Domain
documentationmachine learning
Issue type
documentation
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
2
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
half day
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
Julia fundamentalsFlux.jl basicsanonymous functions
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
80
Research direction
The issue requests expanding Flux.jl's saving/loading documentation to cover anonymous functions. The existing docs likely use JLD2.jl, which fails for anonymous functions. The answer is to use Flux.state and Flux.loadmodel! with explicit assignment of anonymous functions. For cases where the anonymous function holds state, suggest converting to explicit structs. Look at the current docs and any related PRs or comments for specific file locations.