How to count in a one hot tensor

Question

I know I can convert a Tensor to the one-hot using this command:

one_hot_labels = tf.one_hot(labels,depth=3)

Now I want to count how many of class 0, class 1, and class 2 are there in the one_hot_labels. What is the easiest way to count that?

Example:

Input:

one_hot_labels = [[1,0,0],[1,0,0],[0,0,1]]
one_hot_labels.count([1,0,0]) # something like this command

Output:

2

Answer 1

Something like this should work for you:

one_hot_labels = np.array([[1,0,0],[1,0,0],[0,0,1]])
count_label = tf.reduce_sum(one_hot_labels, axis=0)
sess = tf.Session()
sess.run(count_label)
# array([2, 0, 1])

Now for example you can just do:

count_label = tf.reduce_sum(one_hot_labels, axis=0)[0]
# 2