Onboard to the GitHub workflow based issue-labeler

[jeffhandley]

We have a new implementation of dotnet/issue-labeler available that is implemented entirely using GitHub workflows. This approach allows for self-service onboarding and re-training of the prediction models. Onboarding instructions are documented and anyone with write permission to this repository should be able to work through the process of the initial model training, but @RussKie and I are both familiar with the process.

I ran a test set of training from this repository into my fork. The workflow run can be seen here, and the model's accuracy looks like it'll start with a good baseline. It took less than 12 minutes to train the model.

• Issues tested: 2834
  • The predicted label matched the existing label: 2706 (95.48 %)
  • The predicted label does not match the existing label: 46 (1.62 %)
  • No prediction was made: 82 (2.89 %)
  • A prediction was made, but no existing label was present: 0 (0.00 %)
• Pulls tested: 2885
  • The predicted label matched the existing label: 2795 (96.88 %)
  • The predicted label does not match the existing label: : 74 (2.56 %)
  • No prediction was made: 16 (0.55 %)
  • A prediction was made, but no existing label was present: 0 (0.00 %)

[Copilot]

PR Overview

This PR introduces a new GitHub workflow-based implementation of the issue-labeler with self-service onboarding and retraining of prediction models. Key changes include the addition of workflows for training, caching, predicting for issues and pull requests, promoting models, and building the predictor app.

Reviewed Changes

File	Description
.github/workflows/labeler-train.yml	Defines training parameters and triggers using workflow_dispatch inputs
.github/workflows/labeler-cache-retention.yml	Schedules cache retention tasks with a conditional run on dotnet repos
.github/workflows/labeler-predict-pulls.yml	Sets up pull request label prediction with conditional logic
.github/workflows/labeler-predict-issues.yml	Configures issue label prediction with workflow_dispatch inputs
.github/workflows/labeler-promote.yml	Adds workflow for promoting staged models to the live environment
.github/workflows/labeler-build-predictor.yml	Provides a build workflow for the predictor app

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

[danmoseley]

yay! thanks

[jeffhandley]

I fired off the job to train the models. I set it to train directly into the LIVE cache entries to take effect immediately upon completion.

https://github.com/dotnet/aspire/actions/runs/13742535771

[jeffhandley]

Training took less than 10 minutes and all is working. New issues and pull requests will be labeled. If you want to have it backfill existing issue/pull labels, you can manually trigger the respective jobs. I'd expect it to be able to label everything in history in 30 minutes or less.

The FAQ · dotnet/issue-labeler Wiki has notes about when you would need to retrain.

[danmoseley]

seems to be working backfilling.