Home Original page

Fix documentation on softmax_cross_entropy_with_logits_v2 by mbrio · Pull Request #21517 · tensorflow/tensorflow

import tensorflow as tf
import numpy as np

logits = np.array([
    [5.0,  2.3, 0.0, -3.2, -0.8],
    [2.3,  0.1, 1.8,  1.1,  1.2],
    [-1.2, 1.1, 0.1,  2.5,  0.7]
])

labels = np.array([
    [1., 1., 0., 0., 0.],
    [0., 0., 1., 0., 1.],
    [0., 0., 0., 1., 0.]
])

x_entropy = tf.nn.softmax_cross_entropy_with_logits_v2(labels=labels, logits=logits)

with tf.Session() as sess:
    print(sess.run(tf.shape(labels)))  # [3, 5]
    print(sess.run(tf.shape(x_entropy)))  # [3]

In this example we see that the x_entropy shape does not equal the shape of labels as stated in the original documentation, it is of shape [batch_size].