TensorFlow tf.keras.losses.BinaryCrossentropy

cross-entropy loss
$$L=-[ylog\hat{y}+(1-y)log(1-\hat{y})]$$
y为真实值，$\hat{y}$为预测，交叉熵损失函数可以衡量y与$\hat{y}$的相似性。

cross-entropy loss用于二分类问题(假设是0,1),最后一层要使用sigmoid函数配合,输出值使用one-hot编码.(此外Categorical cross-entropy配合softmax层用于多分类问题)

下例中有4个样本,y_pred和y_true都有[batch_size]

bce = tf.keras.losses.BinaryCrossentropy()
loss = bce([0., 0., 1., 1.], [1., 1., 1., 0.])
print('Loss: ', loss.numpy())  # Loss: 11.522857

init

__init__(
    from_logits=False,
    label_smoothing=0,
    reduction=losses_utils.ReductionV2.AUTO,
    name='binary_crossentropy'
)

参数	描述
from_logits
label_smoothing
reduction
name

call

__call__(
    y_true,
    y_pred,
    sample_weight=None
)

参数	描述
y_true
y_pred
sample_weight

参考:
https://blog.csdn.net/koreyoshichen/article/details/84823636
https://blog.csdn.net/zwqjoy/article/details/78952087
https://blog.csdn.net/red_stone1/article/details/80735068