time-series-classification

We should have a shorthand for the reducer where y is simply predicted from X, possibly including the index.

This is generally not a good forecaster in isolation, but useful as a component for instance in #1279, in anomaly detection or other annotation tasks.

It's also a special case of the direct regression strategy with exogeneous variables, where there are zero lags.

Question for @

Hi, Thanks for the awesome library!

So I am running a Kmeans on lots of different datasets, which all have roughly four shapes, so I initialize with those shapes and it works well, except for just a few times. There are a few datasets that look different enough that I end up with empty clusters and the algorithm just hangs ("Resumed because of empty cluster" again and again).

I conceptually

Dear team,

I am in stuck when convert very large numpy array to your TSDatasets.
These are what I have tried to fix my issue:

• when building time-series, I used tensorflow.keras.preprocessing.timeseries_dataset_from_array. After this step, the memory is still fine
• I concatenate all batch data into numpy array, this step produces problem so I use numpy memmap to avoi

comments in sktime-dl are currently in a different style to sktime. Standardising to conform would be helpful. See, for example, this for style.

https://github.com/alan-turing-institute/sktime/tree/main/sktime/classification