CV in time series?

[irahulroy]

My question is more on the theoretical side. I see that the 'model.fit' allows to set CV during algorithm training. Does the CV include random sampling and reshuffling during the training phase, or is it more like rolling-origin cross-validation, where the time chronology is maintained?

Thanks,
Rahul

[ngupta23]

@Rroy09 Random sampling and shuffling is not the the recommended method for CV in time series due to auto correlation issues. We use temporal cross validation similar to what is described here.

[AutoViML]

Yes agreed. No random sampling is done. We do expanding window cross validation. You can think of it as a time window that is constantly expanding to the right. Hope that helps Ram

…

On Thu, Dec 31, 2020 at 8:20 AM Nikhil Gupta ***@***.***> wrote: @Rroy09 <https://github.com/rroy09> Random sampling and shuffling is not the the recommended method for CV in time series due to auto correlation issues. We use temporal cross validation similar to what is described here <https://robjhyndman.com/hyndsight/tscv/>. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#35 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMKBH6EP2QHWAPW5RPFWHXDSXR3CJANCNFSM4VPF4W6A> .

[irahulroy]

Yeah, thanks for the reply. Just wanted to confirm as random sampling is prohibited for time series data.