如何使用自定义损失功能保存和加载Keras模型

Question

我在我的keras模型中编写了一个自定义损失函数

def sparse_cross_entropy(y_true, y_pred):
        """
        Calculate the cross-entropy loss between y_true and y_pred.

        y_true is a 2-rank tensor with the desired output.
        The shape is [batch_size, sequence_length] and it
        contains sequences of integer-tokens.

        y_pred is the decoder's output which is a 3-rank tensor
        with shape [batch_size, sequence_length, num_words]
        so that for each sequence in the batch there is a one-hot
        encoded array of length num_words.
        """

        # Calculate the loss. This outputs a
        # 2-rank tensor of shape [batch_size, sequence_length]
        loss = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y_true,
                                                              logits=y_pred)

        # Keras may reduce this across the first axis (the batch)
        # but the semantics are unclear, so to be sure we use
        # the loss across the entire 2-rank tensor, we reduce it
        # to a single scalar with the mean function.
        loss_mean = tf.reduce_mean(loss)

        return loss_mean

但是在.h5文件中保存和放置模型时，它表明“ sparse_cross_entropy”未定义