Skip to content

Home

Jump to bottom
Jeff Handley edited this page May 1, 2025 · 4 revisions

The Issue Labeler comprises several GitHub Actions and backing applications that orchestrate the process of training a repository's model, testing the model against existing data, and automatically predicting labels for issues and PRs.

Download

Download GitHub issues or pull requests and cache the data. This action downloads data from GitHub and creates tab-separated (.tsv) data files to be consumed by the Train action.

uses: dotnet/issue-labeler/download@{commit-sha}

Inputs

  • type (required): The type of data to download. Must be either 'issues' or 'pulls'.
  • label_prefix (required): The label prefix to be used for model training. Must end in a non-alphanumeric character.
  • excluded_authors: Comma-separated list of authors to exclude.
  • limit: Max number of items to download (newest items are used). Defaults to the max number of pages times the page size.
  • page_size: Number of items per page in GitHub API requests. Defaults to 100 for issues, 25 for pull requests.
  • page_limit: Maximum number of pages to retrieve. Defaults to 1000 for issues, 4000 for pull requests.
  • retries: Comma-separated list of retry delays in seconds. Defaults to '30,30,300,300,3000,3000'.
  • repository: The org/repo to download data from. Defaults to current repository.
  • cache_key: The cache key suffix to use for saving data. Defaults to 'staged'.

Train

Train the Issues or Pull Requests model for label prediction. This action loads the tab-separated issue and pull request data from the Download action, and trains ML.NET models over the data to prepare for making label predictions.

uses: dotnet/issue-labeler/train@{commit-sha}

Inputs

  • type (required): The model to train. Must be either 'issues' or 'pulls'.
  • data_cache_key: The cache key suffix to use for the downloaded data. Defaults to 'staged'.
  • model_cache_key: The cache key suffix to use for the trained model. Defaults to 'staged'.

Test

Test predictions against the Issues and/or Pull Requests model by downloading data and comparing predictions against existing labels.

uses: dotnet/issue-labeler/test@{commit-sha}

Inputs

  • type (required): The model to test. Must be either 'issues' or 'pulls'.
  • label_prefix (required): The label prefix to use for model training. Must end with a non-alphanumeric character.
  • threshold (required): The minimum confidence score for a label prediction. Must be a number between 0.00 and 1.00. The recommended value is 0.40.
  • excluded_authors: A comma-separated list of authors to exclude.
  • limit: The maximum number of items to download. The newest items are downloaded.
  • page_size: The number of items per page in GitHub API requests. Defaults to 100 for Issues and 25 for Pull Requests.
  • page_limit: The maximum number of pages to retrieve. Defaults to 1000 for Issues and 4000 for Pull Requests.
  • retries: A comma-separated list of retry delays in seconds. Defaults to '30,30,300,300,3000,3000'.
  • cache_key: The cache key suffix to use for saving data. Defaults to 'staged'.
  • repository: The org/repo to download data from. Defaults to the current repository.
  • verbose: Enable verbose output.

Promote

Promote a model from staging to 'ACTIVE', backing up the currently 'ACTIVE' model.

uses: dotnet/issue-labeler/promote@{commit-sha}

Inputs

  • type (required): The model to promote. Must be 'issues' or 'pulls'.
  • staged_key: The suffix for the staged cache entry to promote. Defaults to 'staged'.
  • backup_key: The suffix for the backup cache entry. Defaults to 'backup'.

Restore

Restore a model from cache for label prediction or cache retention.

uses: dotnet/issue-labeler/restore@{commit-sha}

Inputs

  • type (required): The model to restore. Must be 'issues' or 'pulls'.
  • cache_key (required): The cache key suffix to use for loading the model. Defaults to 'ACTIVE'.
  • fail-on-cache-miss: Set to 'true' to fail the job if the model cannot be restored from cache. Defaults to 'false'.
  • quiet: Set to 'true' to suppress output into the GitHub action summary. Defaults to 'false'.

Predict

Predict labels for Issues and Pull Requests using models already restored from cache. This action uses a Docker Container Image published to the GitHub Container Registry to run the predictor app and apply labels automatically, quickly, and securely.

uses: dotnet/issue-labeler/predict@{commit-sha}

Inputs

  • issues (required): Issue Numbers (comma-separated list of ranges).
  • pulls (required): Pull Request Numbers (comma-separated list of ranges).
  • label_prefix: The label prefix used for prediction. Must end with a non-alphanumeric character. Defaults to 'area-'.
  • threshold: The minimum confidence score for a label prediction, as a decimal between 0.00 and 1.00. Defaults to 0.40.
  • default_label: The default label to apply if no prediction meets the threshold. Leave blank for no default label.
  • excluded_authors: Comma-separated list of authors to exclude. Defaults to none.
  • retries: Comma-separated list of retry delays in seconds. Defaults to '30,30,300,300,3000,3000'.
  • test: Run in test mode, outputting predictions without applying labels.
  • verbose: Enable verbose output.
Clone this wiki locally