Forecaster

UpdateRefitsEvery

Categorical in XInsamplePred int insampleExogenous

Refits periodically when update is called.

If update is called with update_params=True and refit_interval or more has elapsed since the last fit, refits the forecaster instead (call to fit).

Refitting is done on (potentially) all data seen so far.

refit_window controls the lookback window on which refitting is done. The refit is carried out on all data in the lookback window cutoff (inclusive, end) to cutoff minus refit_window (exclusive, start), with a default lookback window of all data seen so far.

Schnellstart

python

from sktime.forecasting.stream import UpdateRefitsEvery

estimator = UpdateRefitsEvery(forecaster, refit_interval=0, refit_window_size=None, refit_window_lag=0)

Parameter(4)

forecasteran sktime forecaster: the forecaster to be refit/updated regularly
refit_intervaldifference of sktime time indices (int or timedelta), optional: interval that needs to elapse after which the first update defaults to fit default = 0, i.e., always refits, never updates
if index of y seen in fit is integer or y is index-free container type, refit_interval must be int, and is interpreted as difference of int location
if index of y seen in fit is timestamp, must be int or pd.Timedelta
if pd.Timedelta, will be interpreted as time since last refit elapsed
if int, will be interpreted as number of time stamps seen since last refit
refit_window_sizedifference of sktime time indices (int or timedelta), optional: length of the data window to refit to in case update calls fit; default = inf, i.e., refits to entire training data seen so far
refit_window_lagdifference of sktime indices (int or timedelta), optional: lag of the data window to refit to, w.r.t. cutoff, in case update calls fit; default = 0, i.e., refit window ends with and includes cutoff

Beispiele

>>> from sktime.forecasting.trend import TrendForecaster
>>> from sktime.forecasting.stream import UpdateRefitsEvery
>>> from sktime.datasets import load_airline
>>> y = load_airline ()
>>> y0 = y. iloc [: - 20 ]
>>> y1 = y. iloc [- 20: - 10 ]
>>> y2 = y. iloc [- 10:]
>>> forecaster = TrendForecaster ()
>>> forecaster = UpdateRefitsEvery (forecaster, refit_interval = 12)
>>> forecaster. fit (y0, fh = [1, 2, 3 ]) UpdateRefitsEvery(
... )
>>> # predict etc could be called here
>>> # e.g., forecaster.predict()
>>> 
>>> # first update, 10 < refit_interval = 12, so calls update
>>> forecaster. update (y1) UpdateRefitsEvery(
... )
>>> # second update, 20 >= refit_interval = 12, so calls fit
>>> forecaster. update (y2) UpdateRefitsEvery(
... )