Matthew Middlehurst

This seems to be affecting #3090 as well for a range of classifiers and transformers. Most of the failures have been untouched for quite a while as far as I...

~~Note for HIVE-COTE, as it seems all of the individual classifiers are not failing, the issue is most likely in the class itself or the _get_train_probs methods.~~ Disregard, RandomIntervalSpectralEnsemble is...

Possibly a numba issue? Most of them use it in some capacity, not all though, MatrixProfileTransformer uses stumpy (which uses numba https://stumpy.readthedocs.io/en/latest/Tutorial_STUMPY_Basics.html). Very strange, only thing they seem to have...

Was this fixed with the swap to `prefer="threads`? May be worth opening threading issue on the main repo or closing.

Better to discuss threading in `aeon` IMO. this is stale so closing.

Currently running (2) on the 112 datasets. Can run the new datasets when they are ready.

using set_classifier, which does deviate I'm pretty sure

good to compare both either way

Unfortunately, this one is mostly TSFresh. We could try to add a `replace_nans` option to the transformer or FreshPRINCE class for the crashing ones.

That's my assumption based on the error, will try in it in a bit. Either that or the RotationForest transform is giving a nan. Just not much room for optimisation...