patch-transformer supporting multivariate predict univariate

[Essie0715]

Hi Everyone,

I have got a quick question. Does the current patch-transformer support multivariate predict univariate in the current version?

[kdgutier]

Hey @Essie0715,

There are two things you can do for a multivariate PatchTST:

1. You can adapt PatchTST to have multivariate outputs (not inputs) with the Mixture Mesh probabilities.
2. You can feed other series temporal data through the exogenous features parameters futr_exog_list, hist_exog_list.

We have one dedicated multivariate model with the StemGNN.

[Essie0715]

Hi @kdgutier

Thank you for your reply. My task is to use multiple time series (historical temperature, wind, humidity, etc) as input features and a Single output forecasting target (predict the future temperature). The original PatchTST is multiple input and multiple output models. But I want to make it a single output model only.

[kdgutier]

In our current version of PatchTST we treat it as a univariate model (single series output).
Exogenous features support for the PatchTST and other Transformers is in our list of priorities.

Here is a tutorial for exogenous variables: https://nixtla.github.io/neuralforecast/examples/exogenous_variables.html