User Guide

Core concepts for working with Twiga, covering the forecaster interface and configuration system through data preparation, statistical analysis, evaluation metrics and time-based backtesting.

Every component — data pipeline, model architecture, CV strategy and conformal calibration — is configured through validated Pydantic dataclasses that are serialisable, diffable and directly wired to Optuna search spaces. Hyperparameter tuning with TPE sampler and Hyperband pruner, expanding and rolling window backtesting, and checkpoint-based resume are first-class features built into the core interface.

• TwigaForecaster & Forecaster Architecture
• Configuration System
• Data Processing Pipeline
• Feature Analysis & Statistical Toolkit
• Evaluation Metrics
• Backtesting & Time-Based Cross-Validation
• Feature Attribution (SHAP)
• Visualization