Tutorials

End-to-end notebooks that walk through every major feature of Twiga, from loading data to deploying a production MLOps pipeline.

The recommended reading order groups notebooks into five stages:

Foundation - data, features, and point forecasting (01–05)

Start here. Covers data loading, forecastability analysis, feature engineering, time-series differencing, and ML point forecasting.

Evaluation & Baselines - measuring and contextualising accuracy (06–07)

Backtesting with time-based cross-validation, then establishing performance benchmarks with the built-in baseline models and skill scores.

Probabilistic Forecasting - intervals and distributions (08–12)

Neural networks, quantile regression (QR + FPQR), parametric distribution heads, and conformal prediction.

Advanced Workflows - production-grade features (13–16)

Hyperparameter tuning with Optuna, ensemble strategies, custom model integration, the typed forecast result API, and SHAP feature attribution.

MLOps - from training to production (17)

End-to-end MLOps workflow: experiment tracking with MLflow, checkpoint versioning, Evidently drift monitoring, FastAPI serving, and Prefect retraining flows.

Foundation

• Getting Started with Twiga
• Forecastability Analysis
• Feature Engineering
• Time Series Differencing
• ML Point Forecasting with Multiple Models

Evaluation & Baselines

• Backtesting & Robust Evaluation
• Baseline Benchmarking

Probabilistic Forecasting

• Neural Network Models
• Quantile Regression
• Parametric Distributions
• Conformal Prediction & Guaranteed Coverage

Advanced Workflows

• Hyperparameter Tuning with Optuna
• Ensemble Strategies
• Custom Models & Extension Points
• Typed Forecast Results
• SHAP Feature Attribution

MLOps

• MLOps: Tracking, Serving & Monitoring