Open MatthewMiddlehurst opened 1 week ago
@MatthewMiddlehurst
I think this should always be raised
if n_channels != self.n_channels_:
raise ValueError(
"The number of channels in the train data does not match the number "
"of channels in the test data"
)
since we have no mechanism for determining if a classifier can work with different number of channels and this
if n_channels != self.n_channels_:
raise ValueError(
"The number of channels in the train data does not match the number "
"of channels in the test data"
)
only if capability:unequal is false.Have I got that right? PR would do the above and strip out the DrCIF checks
Describe the bug
From #1696
The classification base class does not check for a different series length in
predict
than the one used infit
. This should raise an exception unless the classifier can handle unequal length.Possibly impacts other base classes, but have not checked.
The below example has two estimators raise an exception like excepted, but this in internal to the specific estimator from the following lines:
Steps/Code to reproduce the bug
Expected results
All classifiers except for KNN which can handle unequal length throw an exception in the base class.
Actual results
Only DrCIF and HC2 (which contains DrCIF) throw exceptions, neither in the base class.
Versions
N/A