Conformal Prediction returns error for dataframes with series of unequal length about statsforecast HOT 6 CLOSED

Oh man, yes you are right. Didn‘t think about that. Now it‘s working fine. Thanks for the fast and helpful response!

from statsforecast.

Hey @baniasbaabe, thanks for using statsforecast. The problem isn't the series lengths but the horizon of your intervals, it should either be 1 or match your forecasting horizon (12). In your example it says 2 but I'm pretty sure the error would be a mismatch between 7 (intervals' h) and 12 (forecast's h).

from statsforecast.

Oh okay, I guess this isn‘t the problem because the error will be raised when calling .fit(). So it doesn‘t care about the forecast horizon of the .predict() method. When the series are from the same length, the problem doesn‘t arise.

from statsforecast.

Can you include the full stacktrace?

from statsforecast.

`ValueError Traceback (most recent call last)
in <cell line: 14>()
12 )
13
---> 14 sf.fit(df)
15 sf.predict(h=12, level=[95])

4 frames
/usr/local/lib/python3.10/dist-packages/statsforecast/core.py in fit(self, df, sort_df, prediction_intervals)
878 self.prepare_fit(df, sort_df)
879 if self.n_jobs == 1:
--> 880 self.fitted = self.ga.fit(models=self.models)
881 else:
882 self.fitted_ = self._fit_parallel()

/usr/local/lib/python3.10/dist-packages/statsforecast/core.py in fit(self, models)
75 for i_model, model in enumerate(models):
76 new_model = model.new()
---> 77 fm[i, i_model] = new_model.fit(y=y, X=X)
78 return fm
79

/usr/local/lib/python3.10/dist-packages/statsforecast/models.py in fit(self, y, X)
362 )
363
--> 364 self._store_cs(y=y, X=X)
365 return self
366

/usr/local/lib/python3.10/dist-packages/statsforecast/models.py in _store_cs(self, y, X)
133 def _store_cs(self, y, X):
134 if self.prediction_intervals is not None:
--> 135 self._cs = self._conformity_scores(y, X)
136
137 def _add_conformal_intervals(self, fcst, y, X, level):

/usr/local/lib/python3.10/dist-packages/statsforecast/models.py in _conformity_scores(self, y, X)
124 )
125 fcst_window = self.forecast(h=h, y=y_train, X=X_train, X_future=X_future) # type: ignore[attr-defined]
--> 126 cs[i_window] = np.abs(fcst_window["mean"] - y_test)
127 return cs
128

ValueError: operands could not be broadcast together with shapes (7,) (2,)`

from statsforecast.

Ah, in this case your series are too short. You're asking for 3 windows of size 7, which requires at least 22 samples, but you probably have series with less points than that. You can either remove those series or adjust the intervals settings.

from statsforecast.