loss = y_true * log(y_true / y_pred)
情况一:算两个一维数组
参考:tf.keras.losses.KLDivergence
k = tf.keras.losses.KLDivergence()
loss = k([.4, .9, .2], [.5, .8, .12])
print('Loss: ', loss.numpy()) # Loss: 0.11891246
情况二:按batch来算
参考:TensorFlow->API->TensorFlow Core v2.2.0->Python
import tensorflow as tf
y_true = [[0.3, 0.7], [0.2, 0.8]]
y_pred = [[0.4, 0.6], [0.1, 0.9]]
kl = tf.keras.losses.KLDivergence()
kl(y_true, y_pred).numpy()
实际运算为:
因为这里batch_size=2,所有要求batch维度上的平均。
完整示例: