How to calculate BinaryCrossEntropy loss in TensorFlow

TensorFlow BinaryCrossEntropy

Binary Cross Entropy loss is used when there are only two label classes, for example in cats and dogs image classification there are only two classes i.e cat or dog, in this case Binary Cross Entropy loss can be used. tf.keras api provides implementation of BinaryCrossEntropy, lets understand this with below code snippet.

Create two examples for actual values and predicted values


import tensorflow as tf

actual_values = [[0, 1, 0, 0, 0, 0], [0, 1, 0, 0, 0, 0], [0, 1, 0, 0, 0, 0]]

predicted_values = [[.5, .7, .2, .3, .5, .6],[.5, .7, .7, .2, .5, .6], [.5, .7, .2, .8, .2, .1] ]

actual_values comprises of three batch of actual labels, predicted_values are batches of corresponding predicted values.

Instantiate BinaryCrossEntropy object and compute cross-entropy loss


binary_cross_entropy = tf.keras.losses.BinaryCrossentropy()
loss = binary_cross_entropy(actual_values, predicted_values).numpy()
print(loss)

Output


0.53984624

Complete Code


import tensorflow as tf

actual_values = [[0, 1, 0, 0, 0, 0], [0, 1, 0, 0, 0, 0], [0, 1, 0, 0, 0, 0]]
predicted_values = [[.5, .7, .2, .3, .5, .6],[.5, .7, .2, .3, .5, .6], [.5, .7, .2, .3, .5, .6] ]

binary_cross_entropy = tf.keras.losses.BinaryCrossentropy()
print(binary_cross_entropy)

loss = binary_cross_entropy(actual_values, predicted_values).numpy()
print(loss)