Training on multiple timeseries

[aggregated]

Hello,

First of all I would like to thank you for this excellent framework.

A question, some in-built networks like DeepAR are able to be trained multiple times on multiple timeseries. Is this currently possible with HyperTS, and if so, how would you do it? We have a collection of dozens of timeseries, with covariates that behave in a similar way, and our model would benefit greatly from being trained on all of them.

[zhangxjohn]

Thank you for your interest in HpyerTS. Currently, the built-in DeepAR only supports univariate (with covariates) time forecasting task.