Model governance process

#!/bin/bash

# 1. Register a model in MLflow Registry (from training code)
mlflow models register-model \
  --artifact-uri s3://my-bucket/models/linear-regression/v1 \
  --registered-model-name linear-regression-prod

# 2. Define governance stages in MLflow
mlflow models transition-model-version-stage \
  --name linear-regression-prod \
  --version 1 \
  --stage Staging

# 3. DVC pipeline with governance gates (dvc.yaml)
stages:
  train:
    cmd: python train.py
    deps:
      - train.py
      - data/train.csv
    params:
      - model.learning_rate
      - model.max_depth
    metrics:
      - metrics.json:
          cache: false
    artifacts:
      - models/model.pkl:
          type: model
  
  validate:
    cmd: python validate.py --model models/model.pkl --threshold 0.85
    deps:
      - validate.py
      - data/test.csv
      - models/model.pkl
    metrics:
      - validation.json:
          cache: false

# 4. GitHub Actions workflow for approval gate
name: Model Approval Gate
on:
  workflow_dispatch:
    inputs:
      model_version:
        description: Model version to promote
        required: true
        type: string
jobs:
  approval:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Set up Python
        uses: actions/setup-python@v5
        with:
          python-version: '3.11'
      - name: Install MLflow
        run: pip install mlflow boto3
      - name: Validate model metrics
        run: |
          python -c "
          import mlflow
          from mlflow.tracking import MlflowClient
          client = MlflowClient()
          runs = client.search_runs('0', max_results=1)
          metrics = runs[0].data.metrics
          if metrics.get('accuracy', 0) < 0.85:
            raise ValueError('Model accuracy below threshold')
          print('✓ Model validated')
          "
      - name: Promote to Production
        run: |
          mlflow models transition-model-version-stage \
            --name linear-regression-prod \
            --version ${{ github.event.inputs.model_version }} \
            --stage Production
      - name: Log audit entry
        run: |
          echo "Model ${{ github.event.inputs.model_version }} promoted by ${{ github.actor }} at $(date)" >> AUDIT.log
      - name: Commit audit log
        run: |
          git config user.email "governance@example.com"
          git config user.name "Governance Bot"
          git add AUDIT.log
          git commit -m "Audit: Promote model v${{ github.event.inputs.model_version }}"
          git push

# 5. MLflow config for multi-stage governance (.mlflow/config.yaml)
backend_store:
  uri: postgresql://user:password@localhost/mlflow
default_artifact_root: s3://my-bucket/mlflow-artifacts
model_registry:
  stages:
    - name: Development
      description: Experimental models
      requires_approval: false
    - name: Staging
      description: Validated candidates
      requires_approval: false
    - name: Production
      description: Live models
      requires_approval: true
      approvers:
        - data-science-lead
        - mlops-engineer