Thank you for your answer.
I don’t think the logic is right the way you suggested, or at least I don’t get the logic :D.
If I compute theta with
theta = pm.invlogit((at[:, :, None] * xt_theano).sum(axis=1) + beta)
I get a result of the size 100, 3000. The dot product values of the features at every time step t. But in the last line of the Model
like = pm.Poisson("like", (100-t)*theta, observed=labels)
the label is, for example, for time step 30, so it should only count for the 30th entry of theta. But with the Model right now the label would impact all time steps, entries of theta, from 0-99 correct?
Thats why I originally wanted to receive scalar values for theta