time series classification

This request has been here, and there is a workaround with tsfresh - but that's not in the spirit of autogluon it seems :)
Since you have very powerful transformations for DirectTabular and RecursiveTabular, you may implement it in autogluon.timeseries by recognizing the target as classification and directly switching to train only the tabular ts-models with problem type set to classification.
just for clarification: can this be done right now manually? specifying distinct models is easy with hyperparams, but can I force them to do strictly classification? The manual says no : "TabularPredictor will always be trained with "regression" problem type, (...)"

Start by reading the autogluon.timeseries integration with DirectTabular and RecursiveTabular, then check how TabularPredictor selects its problem type. Confirm whether classification can currently be forced manually; if not, define how target recognition should switch to classification and what behavior demonstrates completion.

Written by the indexing model from the issue text.