解决使用keras提供的损失函数API时，梯度无法反向传播, 损失函数不下降

在使用keras提供的损失函数API时，梯度无法反向传播, 损失函数不下降

问题：

在使用keras提供的损失函数API时，梯度无法反向传播

代码：

from tensorflow.keras.losses import categorical_crossentropy

def train_generator(x, y, z, eps, dcgan, siamese_model, loss=None):

    with tf.GradientTape(persistent=True) as t:
        fake_x = dcgan.generator([z, y])
        loss_G = -tf.reduce_mean(dcgan.discriminator(fake_x)) 
        preds = aux_model(fake_x)
        aux_mean = categorical_crossentropy(y, preds)
        aux_loss = tf.reduce_mean(aux_mean)
        total_loss  = aux_loss + loss_G
        gradient_g = t.gradient(total_loss, dcgan.generator.trainable_variables)

    dcgan.optimizer_G.apply_gradients(zip(gradient_g, dcgan.generator.trainable_variables))

猜测原因：

Keras接口有时候会先对数据进行预处理，然后再调用tensorflow的backend，这样会导致函数梯度链断开，无法通过链式求导来进行梯度下降

查看keras源码是怎么定义categorical_crossentropy的：

root@Ie1c58c4ee0020126c:~# find / -iname keras
/usr/local/lib/python3.7/dist-packages/keras
/usr/local/lib/python3.7/dist-packages/tensorflow_core/contrib/keras
/usr/local/lib/python3.7/dist-packages/tensorflow_core/contrib/keras/api/keras
/usr/local/lib/python3.7/dist-packages/tensorflow_core/python/keras
/usr/local/lib/python3.7/dist-packages/tensorflow_core/python/keras/api/_v1/keras
/usr/local/lib/python3.7/dist-packages/tensorflow_core/python/keras/api/_v2/keras
/usr/local/lib/python3.7/dist-packages/tensorflow_core/python/keras/api/keras

vim /usr/local/lib/python3.7/dist-packages/tensorflow_core/python/keras/losses.py

losses.py

def categorical_crossentropy(y_true,
                             y_pred,
                             from_logits=False,
                             label_smoothing=0):
  """Computes the categorical crossentropy loss.

  Args:
    y_true: tensor of true targets.
    y_pred: tensor of predicted targets.
    from_logits: Whether `y_pred` is expected to be a logits tensor. By default,
      we assume that `y_pred` encodes a probability distribution.
    label_smoothing: Float in [0, 1]. If > `0` then smooth the labels.

  Returns:
    Categorical crossentropy loss value.
  """
  y_pred = ops.convert_to_tensor(y_pred)
  y_true = math_ops.cast(y_true, y_pred.dtype)
  label_smoothing = ops.convert_to_tensor(label_smoothing, dtype=K.floatx())
  def _smooth_labels():
    num_classes = math_ops.cast(array_ops.shape(y_true)[1], y_pred.dtype)
    return y_true * (1.0 - label_smoothing) + (label_smoothing / num_classes)

  y_true = smart_cond.smart_cond(label_smoothing,
                                 _smooth_labels, lambda: y_true)
  return K.categorical_crossentropy(y_true, y_pred, from_logits=from_logits)

可以看到在keras的接口里面对数据做了一些预处理然后再调用了tensorflow backend(K)的categorical_crossentropy接口，所以导致了梯度链的断裂，无法通过链式求导和反向传播了更新梯度

解决方法：

自己实现损失函数，或者取keras的losses.py文件中找到源码，直接调用tensorflow backend提供的接口作为损失函数。

最终将代码改为：

from tensorflow.keras import backend as K

def train_generator(x, y, z, eps, dcgan, siamese_model, loss=None):

    with tf.GradientTape(persistent=True) as t:
        fake_x = dcgan.generator([z, y])
        loss_G = -tf.reduce_mean(dcgan.discriminator(fake_x)) 
        preds = aux_model(fake_x)
        aux_mean = K.categorical_crossentropy(y, preds)
        aux_loss = tf.reduce_mean(aux_mean)
        total_loss  = aux_loss + loss_G
        gradient_g = t.gradient(total_loss, dcgan.generator.trainable_variables)

    dcgan.optimizer_G.apply_gradients(zip(gradient_g, dcgan.generator.trainable_variables))

问题完美解决， 相同问题的可以参考下。