Skip to main content

MLOps

End-to-end machine learning operations including AutoML, experiment tracking, model registry, and deployment pipelines.

AutoML​

Supported Tasks​

  • Classification: Binary and multi-class classification
  • Regression: Numerical prediction
  • Time Series: Forecasting and anomaly detection
  • Clustering: Unsupervised grouping

Features​

  • Automatic feature engineering
  • Hyperparameter optimization
  • Model selection and ensembling
  • Cross-validation and evaluation

Experiment Tracking​

Track all training runs with comprehensive metrics and artifact logging:

  • Metrics: Loss, accuracy, precision, recall, F1, custom metrics
  • Parameters: Hyperparameters, model config, dataset versions
  • Artifacts: Model weights, plots, feature importance, predictions
  • Comparisons: Side-by-side experiment comparison, metric charts
  • Reproducibility: Environment snapshots, random seeds, data versioning

Model Registry​

Version Control for Models​

  • Model Versioning: Track all model versions with metadata
  • Model Cards: Documentation, performance metrics, bias analysis

A/B Testing Routers​

Compare model variants in production with traffic splitting:

  • Weighted Random: Split traffic by percentage between variants
  • Feature-Based: Route based on input features for segment testing
  • Multi-Armed Bandit: Auto-optimize traffic to best performers

Learn more about A/B Testing →

Drift Detection​

Monitor model performance and detect data distribution changes:

  • PSI Monitoring: Population Stability Index for drift detection
  • Feature Drift: Per-feature drift analysis with statistical tests
  • Ground Truth: Upload actual outcomes to track accuracy over time
  • Alerts: Automated notifications when drift exceeds thresholds

Learn more about Drift Detection →

Deployment Pipelines​

# Example: Deploy model from registry
from strongly import MLOps

mlops = MLOps()

# Get production model
model = mlops.get_model('customer-churn', stage='production')

# Deploy as API endpoint
mlops.deploy(
    model_id=model.id,
    endpoint_name='churn-prediction',
    instances=2,
    cpu='500m',
    memory='1Gi'
)

# Monitor predictions
mlops.enable_monitoring(
    endpoint='churn-prediction',
    log_predictions=True,
    detect_drift=True
)