[BUG] NotFittedError raised in BaseGridSearch when refit is False

Author: aiwalterCreated Aug 10, 2022Updated Sep 18, 2026

Labelsbugmodule:forecasting

Describe the bug related discussion: #3220

grid search is trying to predict on self.best_forecaster_ = self.forecaster.clone().set_params(**self.best_params_) which was not fitted and raises therefore a NotFittedError. The problem is thatevaluate()does not return the last fitted forecaster but it can optionally return the y_train for each forecaster, so we can take that data to fit the forecaster if not refit :)

To Reproduce

python

from sktime.datasets import load_shampoo_sales
from sktime.forecasting.model_selection import (
    ExpandingWindowSplitter,
    ForecastingGridSearchCV,
    ExpandingWindowSplitter)
from sktime.forecasting.naive import NaiveForecaster
y = load_shampoo_sales()
fh = [1,2,3]
cv = ExpandingWindowSplitter(
    start_with_window=True,
    fh=fh)
forecaster = NaiveForecaster()
param_grid = {"strategy" : ["last", "mean", "drift"]}
gscv = ForecastingGridSearchCV(
    forecaster=forecaster,
    param_grid=param_grid,
    cv=cv,
    refit=False)
gscv.fit(y)
gscv.predict(fh=[1,2,3])

Expected behavior If refit is False, then the last y_train from cross-validation should be used as fit data

Additional context

Versions main

Source: sktime/sktime

View original on GitHub View discussion on GitHub