retrain.yml

name: Retrain and Deploy

on:
  push:
    branches: [main]
  workflow_dispatch:
  schedule:
    - cron: "0 2 * * 1"

concurrency:
  group: retrain-${{ github.ref }}
  cancel-in-progress: true

jobs:
  should-retrain:
    runs-on: ubuntu-latest
    outputs:
      retrain: ${{ steps.check.outputs.retrain }}
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - name: Detect data changes
        id: check
        run: |
          git fetch origin main
          if [ "${{ github.event_name }}" = "schedule" ] || [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
            echo "retrain=true" >> $GITHUB_OUTPUT
          else
            if [ "${{ github.event_name }}" = "push" ]; then
              CHANGED=$(git diff --name-only ${{ github.event.before }} HEAD | grep -E "^data/raw/(human|ai)/" || true)
            else
              CHANGED=$(git diff --name-only origin/main HEAD | grep -E "^data/raw/(human|ai)/" || true)
            fi
            if [ -n "$CHANGED" ]; then
              echo "retrain=true" >> $GITHUB_OUTPUT
            else
              echo "retrain=false" >> $GITHUB_OUTPUT
            fi
          fi

  train:
    runs-on: ubuntu-latest
    needs: should-retrain
    if: needs.should-retrain.outputs.retrain == 'true'
    steps:
      - uses: actions/checkout@v4

      - uses: actions/setup-python@v5
        with:
          python-version: "3.10"

      - name: Install dependencies
        run: |
          pip install -r requirements.txt

      - name: Feature engineering
        run: python -m src.features.build_features

      - name: Train model
        run: python -m src.models.train

      - name: Evaluate model
        run: python -m src.models.evaluate

      - name: Upload model artifacts
        uses: actions/upload-artifact@v4
        with:
          name: model-artifacts
          path: |
            data/processed/test_metrics.json
            models/checkpoints/
            models/onnx/
            tokenizer/
            models/registry/
            mlruns/
          retention-days: 30

  deploy:
    runs-on: ubuntu-latest
    needs: train
    if: github.ref == 'refs/heads/main'
    environment: production
    steps:
      - uses: actions/checkout@v4

      - uses: actions/setup-python@v5
        with:
          python-version: "3.10"

      - name: Install dependencies
        run: |
          pip install -r requirements.txt

      - name: Download model artifacts
        uses: actions/download-artifact@v4
        with:
          name: model-artifacts
          path: .

      - name: Check F1 deployment gate
        id: f1-gate
        run: python -m src.models.deploy

      - name: Deploy model
        if: steps.f1-gate.outputs.should_deploy == 'true'
        run: |
          docker build -t human-ai-detect:latest -f docker/Dockerfile .
          docker run -d --name human-ai-detect -p 8080:8080 \
            -v ${{ github.workspace }}/models:/app/models \
            -v ${{ github.workspace }}/tokenizer:/app/tokenizer \
            human-ai-detect:latest
          for i in $(seq 1 30); do
            if curl -sf http://localhost:8080/health; then
              echo "Health check passed"
              break
            fi
            echo "Health check attempt $i failed, retrying..."
            sleep 2
          done
          RESPONSE=$(curl -sf -X POST http://localhost:8080/predict \
            -H "Content-Type: application/json" \
            -d '{"code": "#include <stdio.h>\\nint main() { return 0; }"}')
          echo "Human code prediction result: $RESPONSE"
          echo "$RESPONSE" | python3 -c "import sys,json; d=json.load(sys.stdin); assert d['prediction'] == 'human', f\"Expected 'human' but got '{d['prediction']}'\"; assert d['human_probability'] > d['ai_probability'], 'Human probability should be higher for human code'; print('Smoke test passed: human code correctly classified as human')"
          docker stop human-ai-detect
          docker rm human-ai-detect

      - name: Promote model to Production
        if: steps.f1-gate.outputs.should_deploy == 'true'
        run: python -m src.models.registry --promote-staging

      - name: Skip deployment notification
        if: steps.f1-gate.outputs.should_deploy != 'true'
        run: |
          echo "Deployment skipped: ${{ steps.f1-gate.outputs.deploy_reason }}"