Multiple response model, broadcast failure?

Questions
8

My confusion then becomes, why does this logic not extend to the binary response case? If you see the first code block z_student has shape (1,students) while z_question has shape (questions,1) .

OK, I think I see - if you have shapes (1, m) and (n, 1), then numpy will do the broadcasting for you:

>>> a = np.random.randn(1, 5)
>>> b = np.random.randn(3, 1)
>>> (a - b).shape
(3, 5)

On the other hand, if you have shapes (k, m) and (n, k), then it won’t:

>>> a = np.random.randn(2, 5)
>>> b = np.random.randn(3, 2)
>>> (a - b).shape
ValueError: operands could not be broadcast together with shapes (2,5) (3,2)

So perhaps first figure out what shape you expect a-b to be, then reshape them accordingly so that the subtraction works