Hi <!channel>, we are happy to announce that we added a full implementation of the classic DeepAR model (and its

auto

counterpart) from Amazon to the library! The main difference between DeepAR and the standard

LSTM

is the recursive Monte Carlo sampling to produce probabilistic forecasts. Give it a try and let us know how it works!

this is on 1.6.1?

yes!

awesome 👍

This is amazing! Thanks!

@Cristian (Nixtla) Is there a summary page in the documentation that shows which algorithms support exogenous variables, static exogenous variables, etc.? Something like this: https://unit8co.github.io/darts/userguide/covariates.html#global-forecasting-models-gfms

Not right now, but thanks for the suggestion, we will definitely add it to this page: https://nixtla.github.io/neuralforecast/examples/models_intro.html