Np.stack not stacking tensors

I am translating old tensorflow code here, and it fails

    N_x_logit0 = tt.tensordot(X, beta0, axes=(2, 1)) + theta[0]
    N_x_logit1 = tt.tensordot(X, beta1, axes=(2, 1)) + theta[1]
    N_x_logit = np.stack([N_x_logit0, N_x_logit1], axis=-1)    

N_x_logit0 is (1000,80,2) as verified by N_x_logit0.tag.test_value.shape. After np.stack, N_x_logit is (2,) according to the eclipse debugger. N_x_logit does not have tag as an attribute, so I can not check its runtime value. I want to make sure it becomes the shape (1000, 80, 2, 2). Is there any way to do that?

There is theano.tensor.stack and theano.tensor.concatenate.

I recommand you to read/search the for theano operations in the beginning.

I believe stack is what I want. The result should have one more dimension than each member.

I read the np.stack documentation in detail. If I understand correctly, theano.tensor calls is one to one mapped to np calls?

I am really just looking for ways to verify the shape of the stacked result. it seems what the debugger showed is only the symbolic shape, and there should be another way to verify the actual variable shape?

Did you try to follow the code snippet in the theano documentation? It seems quite comprehensive.