Combine ForecastX and TransformedTargetForecaster

[akmalmvic0312]

Hello!

How to combine between ForecastX and TransformedTargetForecaster? I want to predict exogenous variable using moving average in ForecastX and run the some method using TransformedTargetForecaster.

[fkiraly]

You just nest them? Let us know if that does not work for you.

Our recent tutorial from pycon Prague has some examples of pipeline building in notebooks 3 and 4:
https://github.com/sktime/sktime-workshop-pydata-prague-2023
(the graphical pipeline requires 0.22.1 or higher)

Some code copied from there:

from sktime.forecasting.arima import AutoARIMA
from sktime.forecasting.compose import ForecastX
from sktime.forecasting.var import VAR

forecastx = ForecastX(
    forecaster_y=AutoARIMA(sp=1, suppress_warnings=True),
    forecaster_X=VAR(),
)

To combine with a TransformedTargetForecaster, wrap the forecastx above in it, e.g.,

from sktime.transformations.series.difference import Differencer

forecastx_and_ttf = Differencer() * forecaster

# or more verbose
forecast_x_and_ttf = TransformedTargetForecaster([
    ("differencer", Differencer()),
    ("forecastx", forecastx),
])

[akmalmvic0312]

Yes i nest them and try to write this script:

# Perform Grid search with KFold cross validation to find optimal hyperparameter
pipe = TransformedTargetForecaster(steps=[
    ("imputer", Imputer(method="mean")),
    ("minmax", TabularToSeriesAdaptor(MinMaxScaler((0, 1)))),
    ("scaler", TabularToSeriesAdaptor(RobustScaler())),
    ("forecaster", ForecastX(forecaster_X=VAR(), forecaster_y=make_reduction(svm.SVR(), window_length=48, strategy="recursive"), columns=["SMA3", "SMA2"],)),   # Initially window_length=24
])

cv_svr = SingleWindowSplitter(fh=fh_int)

gscv_svr = ForecastingGridSearchCV(
    forecaster=pipe,
    param_grid=param_grid,
    cv=cv_svr,
    n_jobs=-1,
    verbose=1,
    scoring=MeanSquaredError(square_root=True),  # set custom scoring function
    error_score='raise',
    refit=True
)

svr_fit = gscv_svr.fit(y=y_train, X=X_train, fh=fh) #, X_train

However, the results from the script still tend to be stagnant. Is there anything I might have missed in the method I've been using?