MLOps & CI/CD for AI

import mlflow
import mlflow.sklearn
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import f1_score

# Start local MLflow server: `mlflow ui`
mlflow.set_experiment("fraud-detection-v2")

with mlflow.start_run(run_name="rf-100-trees"):
    # Log hyperparameters
    mlflow.log_params({
        "n_estimators": 100,
        "max_depth": 10,
        "class_weight": "balanced"
    })

    model = RandomForestClassifier(n_estimators=100, max_depth=10)
    model.fit(X_train, y_train)

    # Log metrics
    f1 = f1_score(y_test, model.predict(X_test))
    mlflow.log_metric("f1_score", f1)

    # Save model artefact
    mlflow.sklearn.log_model(model, "model")
    print(f"Run ID: {mlflow.active_run().info.run_id}")

# Initialise DVC in your git repo
git init && dvc init

# Track a large dataset (stored in S3, version in git)
dvc add data/training_data.csv
git add data/training_data.csv.dvc .gitignore
git commit -m "Add training data v1"

# Configure remote storage
dvc remote add -d myremote s3://my-bucket/dvc
dvc push  # Upload data to S3

# On a different machine — get the exact same data
git pull
dvc pull  # Download from S3

name: ML Training Pipeline

on:
  push:
    paths: ['src/**', 'data/*.dvc']  # Trigger on code or data change

jobs:
  train-and-evaluate:
    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 dependencies
      run: pip install -r requirements.txt

    - name: Pull data from DVC remote
      run: dvc pull
      env:
        AWS_ACCESS_KEY_ID: ${{ secrets.AWS_KEY }}
        AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET }}

    - name: Train model
      run: python src/train.py --config configs/prod.yaml

    - name: Evaluate model
      run: |
        python src/evaluate.py
        python src/check_metrics.py --min-f1 0.85  # Fail if below threshold

    - name: Deploy if passed
      if: success()
      run: python src/deploy.py --environment production

from evidently.report import Report
from evidently.metric_preset import DataDriftPreset

# Compare training data distribution to recent production data
report = Report(metrics=[DataDriftPreset()])
report.run(reference_data=train_df, current_data=production_last_7_days)

# Save HTML report
report.save_html("drift_report.html")

# Check programmatically
result = report.as_dict()
if result['metrics'][0]['result']['dataset_drift']:
    trigger_retraining_pipeline()