search

awslabs / gluonts

Probabilistic time series modeling in Python
https://ts.gluon.ai
Apache License 2.0
4.51k stars 743 forks source link

Available estimators for high dimension multivariable time series? #692

Open wangzhennnn opened 4 years ago

wangzhennnn commented 4 years ago

Hi, I have a dataset which contains 700-dimensional time series. On this setting, some estimators such as Deepar may not work very well. So my question is which estimators could work very well for high dimension multivariable time series dataset on GluonTS? Another question is that how to set distribution if I have multi-features for each time series? Thank you very much!

StatMixedML commented 4 years ago

On this setting, some estimators such as Deepar may not work very well.

Not sure what you mean, as DeepAR can handle high dimensional time series settings very well. In case you expect the series to be interdependent, you may want to look at:

You find the distribution argument in the help of the functions, distr_output. Choosing an appropriate distribution depends on the support of your response, i.e., count data, real valued y or y>0.