ReconcilerForecaster

Hierarchical reconciliation forecaster.

Quickstart

from sktime.forecasting.reconcile import ReconcilerForecaster

estimator = ReconcilerForecaster(forecaster, method='mint_shrink', return_totals=True, alpha=0)

Parameters(4)

forecasterestimator

Estimator to generate base forecasts which are then reconciled

method{“mint_cov”, “mint_shrink”, “ols”, “wls_var”, “wls_str”, “bu”, “td_fcst”}, default=”mint_shrink”

The reconciliation approach applied to the forecasts based on:

• "mint_cov" - sample covariance

• "mint_shrink" - covariance with shrinkage

• "ols" - ordinary least squares

• "wls_var" - weighted least squares (variance)

• "wls_str" - weighted least squares (structural)

• "bu" - bottom-up

• "td_fcst" - top down based on forecast proportions

return_totalsbool

Whether the predictions returned by predict and predict-like methods should include the total values in the hierarchy, stored at the __total index levels.

• If True, prediction data frames include total values at __total levels

• If False, prediction data frames are returned without __total levels

alpha: float default=0

Optional regularization parameter to avoid singular covariance matrix

Examples

>>> from sktime.forecasting.naive import NaiveForecaster
>>> from sktime.forecasting.reconcile import ReconcilerForecaster
>>> from sktime.transformations.hierarchical.aggregate import Aggregator
>>> from sktime.utils._testing.hierarchical import _bottom_hier_datagen
>>> agg = Aggregator ()
>>> y = _bottom_hier_datagen (
... no_bottom_nodes = 3,
... no_levels = 1,
... random_seed = 123,
... length = 7,
... )
>>> y = agg. fit_transform (y)
>>> forecaster = NaiveForecaster (strategy = "drift")
>>> reconciler = ReconcilerForecaster (forecaster, method = "mint_shrink")
>>> reconciler. fit (y) ReconcilerForecaster(
... )
>>> prds_recon = reconciler. predict (fh = [1 ])

References

1. [1 ] https://otexts.com/fpp3/hierarchical.html