New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[BUG] MSTL documentation is misleading after 0.24.0 release #5454
Comments
Hm, both STL and MSTL should be able to inverse transform in theory. I know the feature was added for STL in one of the recent versions so the |
@fkiraly I have a small typo in the issue. I was playing with different versions and forgot to change from 0.23.0 to 0.24.0. The error logs have changed but the output is still the same that MSTL can not inverse_transform on indices not seen in fit(). I will see the changes in the source code for STLTransformer. Will experiment with it. FitFailedWarning: System: Python dependencies: |
I updated the docstring for #5455. I looked at MSTL, my intention is just to replicate the STL |
This PR adds an inverse transform for the `MSTL` transformer, in line with the `STLTransformer`, which sums up any components from the decomposition. Also updates the documentation, and in the process fixes #5454. Regarding deprecation, the behaviour should be identical in all cases where no bug was present, i.e., identical indices seen in `fit` and `transform`.
Describe the bug
Up until the recent version, MSTL had a note that it can't inverse. The same note is still present in STLTransformer: STLTransformer can not inverse_transform on indices not seen in fit(). This means that for pipelining, the Deseasonalizer or Detrender must be used instead of STLTransformer. This creates a misleading situation for the users who could have assumed that MSTL can inverse_transform.
To Reproduce
FitFailedWarning:
In evaluate, fitting of forecaster TransformedTargetForecaster failed,
you can set error_score='raise' in evaluate to see
the exception message. Fit failed for len(y_train)=15.
The score will be set to nan.
Failed forecaster with parameters: TransformedTargetForecaster(steps=[('imputer', Imputer(method='mean')),
('deseasonalizer', MSTL(periods=[7])),
('forecaster',
NaiveForecaster(strategy='drift'))]).
result = _evaluate_window(
Expected behavior
Ideally, MSTL should have an inverse_transform feature available. Otherwise, the documentation description should change.
Versions
sktime 0.24.0
The text was updated successfully, but these errors were encountered: