trainer.yml

name: Dataset trainer

on:
  workflow_call:
    inputs:
      edit-set-name:
        required: true
        type: string
      train-set-ids:
        required: true
        type: string
      trial-set-ids:
        required: true
        type: string
      bayes-set-ids:
        required: true
        type: string

jobs:
  trainer:
    runs-on: ubuntu-20.04
    continue-on-error: true
    steps:
      - uses: actions/checkout@v2

      - name: Set up Python 3.9
        uses: actions/setup-python@v2
        with: { python-version: '3.9' }

      - name: Install cbng-trainer
        run: pip install git+https://github.com/cluebotng/trainer.git@main

      - name: Download reviewed edits
        run: |
          filter=""
          if [ ! -z "${{ join(fromJSON(inputs.train-set-ids)) }}" ];
          then
            export filter="${filter} --edit-set ${{ join(fromJSON(inputs.train-set-ids), ' --edit-set ') }}"
          fi
          cbng-trainer download-edits --output=train-edits.xml $filter

          # If we have specific edits for the bayes set, then download them. Otherwise, use the same list as train.
          # Note: We duplicate it here, so we can treat the artifact the same in all other steps
          if [ ! -z "${{ join(fromJSON(inputs.bayes-set-ids)) }}" ];
          then
            cbng-trainer download-edits --output=bayes-edits.xml --edit-set ${{ join(fromJSON(inputs.bayes-set-ids), ' --edit-set ') }}
          else
            cp -v train-edits.xml bayes-edits.xml
          fi

      - name: Build databases
        run: |
          mkdir -p "results/"
          cbng-trainer build-database --ann-input=train-edits.xml --bayes-input=bayes-edits.xml --output "results/"

      - name: Prepare files
        run: |
          sudo chown -R runner:docker "results/"
          sudo find results -type f -exec chmod 644 {} \;

          cp train-edits.xml "results/"
          cp bayes-edits.xml "results/"

      - name: Run trial
        run: |
          # If we have specific edits for the trail set, then download them.
          # Otherwise, download a random set of edits to use for comparison.
          if [ ! -z "${{ join(fromJSON(inputs.trial-set-ids)) }}" ];
          then
            cbng-trainer download-edits --output=trial-edits.xml --edit-set ${{ join(fromJSON(inputs.trial-set-ids), ' --edit-set ') }}
          else
            cbng-trainer download-edits --output=trial-edits.xml --random-edits
          fi

          cp trial-edits.xml "results/"

          # Run the actual trial logic
          cbng-trainer trial-database --input trial-edits.xml --output "results/"

      - name: Cleanup artifacts before release
        id: clean
        run: |
          rm -f results/Dockerfile

          # Re-structure the trial output (can't use a directory structure in releases)
          for file in results/trialreport/*;
          do
            file_name=$(basename $(echo $file | cut -d/ -f3))
            cp "$file" "results/trialreport-${file_name}"
          done
          rm -rf results/trialreport/

          echo "::set-output name=date::$(date +'%Y-%m-%d')"

      - name: Crease a release
        uses: ncipollo/release-action@v1
        with:
          token: "${{ secrets.GITHUB_TOKEN }}"
          artifacts: "results/*"
          tag: "${{ inputs.edit-set-name }}/${{ steps.clean.outputs.date }}"