tf.keras.losses.KLDivergence KL散度 损失函数 示例

KL 散度

loss=正确值×log(正确值/预测值)loss = 正确值 \times log(正确值/预测值)loss=正确值×log(正确值/预测值)

import tensorflow as tf

y_true = [[0, 1], [0, 0]]y_pred = [[0.6, 0.4], [0.4, 0.6]]# Using 'auto'/'sum_over_batch_size' reduction type.kl = tf.keras.losses.KLDivergence()kl(y_true, y_pred).numpy()

0.45814306