sktime - HIVECOTEV2 model parameters issue

[monasound]

When I try to run the HIVECOTEV2 model, the following is the error that I encounter. I am using default parameters. I dont understand what the error is and I appreciate your comments.

from sktime.classification.hybrid import HIVECOTEV2
hc2 = HIVECOTEV2(time_limit_in_minutes=1,random_state=1234)
hc2.fit(X_train_Lullymore2 ,y_train_Lullymore)

UserWarning: TemporalDictionaryEnsemble warning: min_window = 10 is larger than max_window = 9. min_window has been set to 9.
warnings.warn(

[fkiraly]

It looks like this comes from an interaction of TDE's defaults min_window and max_window_prop parameters, and your time series being - probably? - of length 9.

The default for max_window is length of your time series (instances), which I assume is 9. And for min_window it is the integer 10.

That should explain the warning?

As a side note, if your time series are very short, you may like to try simply flattening to an n times 9 data frame (instances times timepoits), and applying any sklearn classifier (e.g., RandomForestClassifier) - and use this as a "simple tabular" baseline.

Why: many of the "bespoke" classifiers assume long and/or reasonably smooth time series. Wheras 9 time points usually means very coarse sampling.

The code for a "tabularized" classifier, with sktime interface, would be:

from sklearn.ensemble import RandomForestClassifier
from sktime.transformations.panel.reduce import Tabularizer

clf = Tabularizer() * RandomForestClassifier()

[monasound]

Yes, the one that I use consist of 9. I appreciate your help very much. Thank you for your time.

[fkiraly]

you're welcome! If this solved your problem, can you click "mark as answer" so this question is filed as answered?

[fkiraly]

some additional info:

• in HIVECOTEV2 you can set the parameters of the wrapped TDE via tde_params
• you can build your own HIVE-COTE-like ensembles using WeightedEnsembleClassifier out of your favourite classifiers, if one of the components in the bespoke ensembles (like HIVECOTEVXYZ) keep acting up

[monasound]

Thank you once again