Splitter

ExpandingWindowSplitter

Expanding window splitter.

Split time series repeatedly into an growing training set and a fixed-size test set.

Test window is defined by forecasting horizons relative to the end of the training window. It will contain as many indices as there are forecasting horizons provided to the fh argument. For a forecasating horizon \((h_1,\ldots,h_H)\), the training window will consist of the indices \((k_n+h_1,\ldots,k_n+h_H)\).

For example for initial_window = 5, step_length = 1 and fh = [1, 2, 3] here is a representation of the folds:

|-----------------------|
| * * * * * x x x - - - |
| * * * * * * x x x - - |
| * * * * * * * x x x - |
| * * * * * * * * x x x |

* = training fold.

x = test fold.

Schnellstart

python

from sktime.split.expandingwindow import ExpandingWindowSplitter

estimator = ExpandingWindowSplitter(fh=1, initial_window: int | float | Timedelta | timedelta | timedelta64 | DateOffset=10, step_length: int | Timedelta | timedelta | timedelta64 | DateOffset=1)

Parameter(3)

fhint, list or np.array, optional (default=1): Forecasting horizon
initial_windowint or timedelta or pd.DateOffset, optional (default=10): Window length of initial training fold. If =0, initial training fold is empty.
step_lengthint or timedelta or pd.DateOffset, optional (default=1): Step length between windows

Beispiele

>>> import numpy as np
>>> from sktime.split import ExpandingWindowSplitter
>>> ts = np. arange (10)
>>> splitter = ExpandingWindowSplitter (fh = [2, 4 ], initial_window = 5, step_length = 2)
>>> list (splitter. split (ts)) '[(array([0, 1, 2, 3, 4]), array([6, 8]))]'