update

TimeSeriesPredictor.update(data: TimeSeriesDataFrame | DataFrame | Path | str) → list[str]

[Experimental] Re-select the ensemble on a fresh validation window computed from data.

The most recent validation window from data is appended to the out-of-fold predictions from training (dropping the oldest window) to re-score all models and re-fit the weighted ensemble. Base models are not retrained, so this performs model selection based on more recent data.

Warning

This is experimental and unstable functionality. The API and behavior may change without warning in future releases.

Parameters:

data (TimeSeriesDataFrame | pd.DataFrame | Path | str) – Fresh time series data used to compute the new validation window. Only items (time series) present in both data and the training data are used to re-fit the ensemble.

Returns:

updated_models – Names of the models that were updated.

Return type:

list[str]