ColumnEnsembleForecaster
Forecast each series with separate forecaster.
Applies different forecasters by columns.
ColumnEnsembleForecaster is passed forecaster/index pairs, exact syntax below. Index can be single pandas index element, pd.Index, int, str, or list thereof. If iterable (pd.Index, list), refers to multiple columns.
Behaviour in fit, predict, update: For index pairs f_i, ix_i passed, applies forecaster f_i to column(s) ix_i. predict results are concatenated to one container with same columns as in fit.
Schnellstart
from sktime.forecasting.compose import ColumnEnsembleForecaster
estimator = ColumnEnsembleForecaster(forecasters)Parameter(1)
- forecasterssktime forecaster, or list of tuples (str, estimator, int or pd.index)
if tuples, with name = str, estimator is forecaster, index as int or index
if last element is index, it must be int, str, or pd.Index coercible
if last element is int x, and is not in columns, is interpreted as x-th column
All columns must be present in an index
If forecaster, clones of forecaster are applied to all columns.
If list of tuples, forecaster in tuple is applied to column with int/str index
Beispiele
>>> import pandas as pd
>>> from sktime.forecasting.compose import ColumnEnsembleForecaster
>>> from sktime.forecasting.naive import NaiveForecaster
>>> from sktime.forecasting.trend import PolynomialTrendForecaster
>>> from sktime.datasets import load_longley Using integers (column iloc references) for indexing:
>>> y = load_longley ()[1 ][["GNP", "UNEMP" ]]
>>> forecasters = [
... ("trend", PolynomialTrendForecaster (), 0),
... ("naive", NaiveForecaster (), 1),
... ]
>>> forecaster = ColumnEnsembleForecaster (forecasters = forecasters)
>>> forecaster. fit (y, fh = [1, 2, 3 ]) ColumnEnsembleForecaster(
... )
>>> y_pred = forecaster. predict () Using strings for indexing:
>>> df = pd. DataFrame ({ "a": [1, 2, 3 ], "b": [4, 5, 6 ]})
>>> fc = ColumnEnsembleForecaster (
... [("foo", NaiveForecaster (), "a"), ("bar", NaiveForecaster (), "b")]
... )
>>> fc. fit (df, fh = [1, 42 ]) ColumnEnsembleForecaster(
... )
>>> y_pred = fc. predict () Applying one forecaster to multiple columns, multivariate:
>>> df = pd. DataFrame ({ "a": [1, 2, 3 ], "b": [4, 5, 6 ], "c": [7, 8, 9 ]})
>>> fc = ColumnEnsembleForecaster (
... [("ab", NaiveForecaster (), ["a", 1 ]), ("c", NaiveForecaster (), 2)]
... )
>>> fc. fit (df, fh = [1, 42 ]) ColumnEnsembleForecaster(
... )
>>> y_pred = fc. predict ()