Back to models
Transformer

NonNegativeOptimalReconciler

API referenceSource
Inverse transformUnequal-length series

Apply non-negative reconciliation to hierarchical time series.

Uses all the forecasts to obtain a reconciled forecast, avoiding negative values.

The optimization problem tries to find the bottom forecasts $b$ which are non-negative and minimize the distance between the base forecasts and the reconciled forecasts, given the invertion of error covariance matrix $E$ as a weighting matrix.

Quickstart

python
from sktime.transformations.hierarchical.reconcile import NonNegativeOptimalReconciler

estimator = NonNegativeOptimalReconciler(error_covariance_matrix: DataFrame=None, alpha=0)

Tags

Capabilities

  • Inverse transform: Supported
  • Unequal-length series
  • Multivariate: Not supported
  • Missing values: Not supported
  • Removes missing values: Not supported
  • Equalizes series length: Not supported

Properties

Input typescitype:transform-input
Series
Output typescitype:transform-output
Series
Label typescitype:transform-labels
None
Fit is emptyfit_is_empty
No
Keeps the time indextransform-returns-same-time-index
No
Requires Xrequires_X
Yes
Requires yrequires_y
No
X and y need the same indexX-y-must-have-same-index
No

Parameters(2)

error_covariance_matrixpd.DataFrame, default=None
Error covariance matrix. If None, it is assumed to be the identity matrix
alphafloat, default=0
Constant added to the diagonal of the inverted matrix.

Examples

>>> from sktime.transformations.hierarchical.reconcile import (
... NonNegativeOptimalReconciler)
>>> from sktime.utils._testing.hierarchical import _make_hierarchical
>>> from sktime.forecasting.exp_smoothing import ExponentialSmoothing
>>> y = _make_hierarchical ()
>>> pipe = NonNegativeOptimalReconciler () * ExponentialSmoothing ()
>>> pipe = pipe. fit (y)
>>> y_pred = pipe. predict (fh = [1, 2, 3 ])