future_regressors_model == 'neural_nets' gives me error of the kind: shape '[16, 1, 7]' is invalid for input of size 224

[akosmaroy]

Hi,

When trying to use future regressors with future_regressors_model == 'neural_nets', I'm getting an error message on model.fit() of the following kind:

shape '[16, 1, 7]' is invalid for input of size 224

where 16 x 1 x 7 = 112, which is half of 224

If I change the batch size, I get in essense the same error (the left side will always be half in total than the right side)

Trying to play with future_regressors_d_hidden or future_regressors_num_hidden_layers settings doesn't change this behaviour.

Not adding future regressors gets rid or the error. using future_regressors_model == 'neural_nets' and adding the same future regressors also removes the error.

This is with the current head of the git source code in main, hash 2354356

Whan can I be doing wrong?