univariate time series

Question

univariate time series

tgarriga opened this issue 2 years ago · 6 comments

tgarriga commented 2 years ago

Answer 1 · 2023-02-22T09:56:54.000Z

This model could be used for univariate time series?

Answer 2 · 2023-02-22T09:58:06.000Z

which model exactly? the framework consists of a bunch of models.

Answer 3 · 2023-02-22T10:02:25.000Z

Thanks for your quick response. I mean the temporal conditioned normalizing flows proposed in Multivariate Probabilistic Time Series Forecasting via Conditioned Normalizing Flows
.

Answer 4 · 2023-02-22T10:04:52.000Z

that model is multivariate only, in that it learns the joint distribution over the variates using normalizing flows...

Answer 5 · 2023-02-22T11:08:20.000Z

I understand. First I had read fast the paper and thought that you were conditioning the flow to obtain a fixed size output time series and not just one step, in which case it would make sense to work also with univaritate time series. Have you tried that or know someone who did it?
thanks

Answer 6 · 2023-02-22T11:10:57.000Z

I have not tried it, since there are methods to model the 1-d distribution without specifying some parametric distribution, e.g. the conditional splines methods referenced in the following paper: https://arxiv.org/abs/2107.03743 These methods are implemented in gluonts if you want to use them.

…

On 22. Feb 2023, at 12:08, tgarriga ***@***.***> wrote: I understand. First I had read fast the paper and thought that you were conditioning the flow to obtain a fixed size output time series and not just one step, in which case it would make sense to work also with univaritate time series. Have you tried that or know someone who did it? thanks — Reply to this email directly, view it on GitHub <#126 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAB7JCIISFJXNBXLAX5YH3WYXXS7ANCNFSM6AAAAAAVEDPDOY>. You are receiving this because you commented.