Add the elements in a row, for each row, in a matrix, using theano

Question

I have a vector v and a matrix z, say

v = theano.shared(rng.normal(0, 1, 10))
z = theano.shared(rng.normal(0, 1, (10, 10)))

I want to create a new vector y given by v + sum of elements in each row of z. Basically: y[i] = v[i] + T.sum(z[:,i]) that I can do, for each entry, by:

y[i] = v[i] + theano.tensor.sum(z[:][i])

My question is: is there a way, without doing a loop, to write y = v + T.sum(rows of z) in one single line?

Answer 1

You can obtain it like this

y = v + z.sum(axis=1)

In numpy, and thus in theano, many aggregator functions, such as sum, mean, var, std, any, all, ... have an axis keyword argument, and sometimes even an axes keyword argument, with which you can specify exactly in which direction you array is to be traversed.