Tutorials¶

Step-by-step guides that teach you the fundamentals of time series forecasting with Yohou. Start with the first tutorial below, then pick a path based on your interest.

Start here¶

• Getting Started: Install Yohou, load a dataset, build a full forecasting pipeline, and evaluate multiple models (continuous target). This tutorial gives you the foundation for everything else.

More output types¶

• Class-Probability Forecasting: Forecast categorical outcomes as a probability distribution over classes (categorical target, probability output).
• Interval Forecasting: Produce prediction intervals with statistical coverage guarantees (continuous target, prediction intervals).

Production workflows¶

• Forecasting Workflow: Cross-validation, hyperparameter search, and residual diagnostics (continuous target).
• Observe/Predict Workflow: Step through a test set in batches using the observe/predict loop (continuous target). Read Forecasting Workflow first.
• Exogenous Features: Incorporate external data (X_actual, X_future, X_forecast) into your models (continuous target).
• Cross-Validation Splitters: Create temporal train/test folds with expanding and sliding window strategies. Assumes familiarity with the core pipeline.

Advanced modeling¶

• Reduction Strategies: Compare multi-output, direct, and dir-rec reduction strategies.
• Decomposition: Build a DecompositionPipeline with trend, seasonality, and residual forecasters.
• Panel Data: Forecast multiple related time series simultaneously using the __ naming convention and LocalPanelForecaster. Assumes familiarity with the core pipeline.

Visualization¶

• Exploratory Visualization: Plot rolling statistics, boxplots, missing data, outliers, and resampling comparisons.
• Forecast Visualization: Visualize single and multi-model forecasts, intervals, decomposition, and time weights.
• Seasonal Analysis: Analyze seasonality with overlays, ACF/PACF, STL decomposition, and heatmaps.