[BUG] Number of folds reported in ForecastingGridSearchCV seems to be incorrect

[ngupta23]

Describe the bug
It seems that the number of splits calculation ignores the initial_window. In ForecastingGridSearchCV since the window data gets appended to the initial_window, the number of splits should be far less than the one reported in the example below

To Reproduce

y = load_airline()

fh = np.arange(1, 13)
y_train, y_test = temporal_train_test_split(y, test_size=len(fh))
print(y.shape, y_train.shape[0], y_test.shape[0])

cv = SlidingWindowSplitter(initial_window=int(len(y_train) * 0.5), start_with_window=True)

regressor = RandomForestRegressor(random_state=42)
forecaster = ReducedRegressionForecaster(regressor=regressor, strategy="recursive")

param_grid = {"regressor__n_estimators": [200, 300]}

gscv = ForecastingGridSearchCV(forecaster=forecaster, cv=cv, param_grid=param_grid, verbose=1)
gscv.fit(y_train)

(144,) 132 12
Fitting 122 folds for each of 2 candidates, totalling 244 fits

Expected behavior

With an internal window length of 10 being used, I would have expected the following

Number of folds
= len(validation data) - window_length
= len(y_train) - initial_window - window_length
= 132 - 66 - 10
= 56 folds (expected) but output shows 122 folds.

Below image is for illustration only (does not match the exact number of points in this example)

Versions

System: python: 3.6.12 |Anaconda, Inc.| (default, Sep 9 2020, 00:29:25) [MSC v.1916 64 bit (AMD64)] executable: C:\Users\Nikhil\.conda\envs\sktime_dev\python.exe machine: Windows-10-10.0.18362-SP0

Python dependencies:
pip: 20.3.3
setuptools: 51.0.0.post20201207
sklearn: 0.24.0
sktime: 0.5.1
statsmodels: 0.12.1
numpy: 1.19.4
scipy: 1.5.4
Cython: 0.29.17
pandas: 1.1.5
matplotlib: 3.3.3
joblib: 1.0.0
numba: 0.52.0
pmdarima: 1.8.0
tsfresh: 0.17.0

[mloning]

Yes that's right, good catch! Would appreciate a PR!

[ngupta23]

OK, let me give it a shot and submit a PR.

[mloning]

Closed by #690