Lightning-AI / Lightning-AI/pytorch-lightning

Add Average Callback

Open
#20,617 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

callback feature
Dominant language
Python
Stars
31.4k
Forks
3.8k
Avg merge
6d 7h
Merged PRs (30d)
6

Description

### Description & Motivation

Create a callback to allow creating an average (smoothing) of a logged metrics and log it.
Why directly in lightning ? This feature could be useful for Early Stopping, Now in some training the loss (for example) is very unstable and you could get an exceptional minimum, but because it's very unstable, it's not the best epochs to stop.

Get an log average feature could smooth the loss (for example again) and allow better EarlyStopping, so better training

### Pitch

Create Callback class that get in parameter : the metrics to smooth, the name of the smoothed metric and the callable function to use to smooth the metric, the window, (how many previous values to consider).
The callable function could be mean by default and window set to 5 by default.

### Alternatives

_No response_

### Additional context

_No response_

cc @lantiga @borda

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

The issue names no files, tests, or entry points. Start by locating the callback and metric-logging APIs; done means a configurable smoothing callback logs a named metric from the requested input metric, callable, and window, with mean and a window of 5 as defaults.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
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.