Forecasting Horizon when calling the fit method

[edzWH]

If I'm calling the fit method after doing some CV on a pipeline, is the forecasting horizon irrelevant if I'm purely using the object to document the best params/scores and not fit a model?

For additional context, here is the code I'm running:

` gscv = ForecastingRandomizedSearchCV(forecaster, cv=cv, param_distributions=param_grid, n_iter=random_grid, refit=True)

    result = gscv.fit(y=y_train, X=X_train,fh=fh)`

After the random grid search has finished, I'm not looking to fit the model to future data, I'm just creating a dict out of the results to save elsewhere. As I'm not forecasting in this instance can the forecasting horizon just be a minimum length array? Or should the FH be consistent with the test length in the CV?

Thanks.

[fkiraly]

If I'm calling the fit method after doing some CV on a pipeline, is the forecasting horizon irrelevant if I'm purely using the object to document the best params/scores and not fit a model?

Yes - in this case the fh in fit is unimportant, it is ignored in the internal benchmarking. What matters is the fh in, e.g., SlidingWindowSplitter, this is not automatically set to coincide with the fh you use later for a forecast.

If you only want to run a benchmarking experiment - and not make an additional forecast using the best parameter, you may want to consider using evaluate (from forecasting.model_evaluation), this is also used inside the grid search estimator.