Using UniTS for new tasks
vikolss opened this issue · comments
Great repository! I would like to be able to use the code to pretrain/finetune using custom tasks that do not quite fit into any of the forecast, imputation, anomaly detection or classification tasks. Is it possible (without huge changes to the code) to train/test with other tasks?
Specifically, I would like to be able to do regression, i.e., predict one or more continuous values, that are not future values of the time series itself.
I would also like to be able to to have the output be a sequence of varying length (same length as input sequence), where each element is either a set of continuous values, or a categorical value.
I would greatly appreciate some guidance in this.
Viktor