How many rows are equals between two matrices in numpy

Question

I'm training a neural network for a multi-label classification problem. I have two rows A and B of size BxL (B = batch size, L = number of labels) where A are the actual labels of the minibatch and B are the predictions maded by my model, somethin like that:

A = array([[0., 1., 0.],
          [0., 1., 1.],
          [0., 1., 0.],
          [0., 0., 0.]])

B = array([[1., 1., 0.],
          [0., 1., 1.],
          [0., 0., 0.],
          [0., 1., 0.]])

And I want to count how many samples are classified correctly (that means, how many rows are equals in A and B)

I would like to know if there is a way to do it with tensor/numpy parallel functions...

something like

sum(torch.eq(A,B, axis=0)) # that doesn't exists

Answer 1

You can use numpy with all:

np.sum((A == B).all(1))
#1

This works by searching whether all values in each row match.

>>> A == B
array([[False,  True,  True],
       [ True,  True,  True],
       [ True, False,  True],
       [ True, False,  True]])

Gives you where the elements match, then all(axis=1) returns a boolean of rows where all values are True:

>>> (A == B).all(1)
array([False,  True, False, False])

Showing that the row at index 1 is an exact match between the two arrays.

Then, summing over the boolean array gives you the count of such rows.