feature_spec for cases with multiple response variables

[zavud]

When I have one response variable, it is easy to use feature_spec for data preprocessing:

library(keras)
library(tfdatasets)
spec <- feature_spec(my_training_data, response_variable ~ .)

However, how do we do it for the cases where we have multiple response variables? This stackoverflow post also asks about it, but there is no answer yet.

[t-kalinowski]

Hi, unfortunately this feature is not supported in feature_spec() currently.