CatBoost Time Series
CatBoost trained on lag and calendar features with native categorical handling
CatBoost Time Series uses CatBoost on lag-derived features, with native handling of categorical calendar features (day of week, month, quarter) without manual encoding.
When to use:
- Time series with strong calendar and seasonal patterns (day-of-week effects, holidays)
- When minimal preprocessing of categorical calendar features is preferred
- Datasets with mixed numeric lag features and rich categorical context
Input:
- Trained model checkpoint — exported CatBoost model
- Preprocessing config — lag and feature settings
- Training tail — last N observations
- Steps — forecast horizon
Output: Forecasted values for the specified horizon
Model Settings (set during training, used at inference)
Iterations (default: 1000) Number of boosting rounds.
Learning Rate (default: auto) Step size for gradient updates.
Depth (default: 6) Symmetric tree depth.
Lags (set during training) Historical lag steps included as features.
Inference Settings
No dedicated inference-time settings.