python - "Please wrap your loss computation in a zero argument ` lambda`."是什么意思？

Question

我编写了一个将 L2 损失添加到主要损失函数的代码:

def add_l2(model, penalty=0.001):
    for layer in model.layers:
        if "conv" in layer.name:
            model.add_loss(penalty * tf.reduce_sum(tf.square(layer.trainable_variables[0])))
    return

## training
@tf.function
def train_one_step(model, x, y, optimizer):
    with tf.GradientTape() as tape:
        logits = model(x, training=True)
        loss = _criterion(y_true=y, y_pred=logits)

        add_l2(model, 0.001)
        loss += sum(model.losses)

    grads = tape.gradient(loss, model.trainable_variables)
    optimizer.apply_gradients(zip(grads, model.trainable_variables))
    return loss, logits

当我开始训练时，出现如下错误:

ValueError: Expected a symbolic Tensors or a callable for the loss value. Please wrap your loss computation in a zero argument lambda.

这个错误是什么意思？我该如何治疗？

Answer 1

您的损失引用了模型层之一的变量(layer.trainable_variables[0])，因此需要将您的损失包装在零参数 lambda 中以使其成为可调用的。model.add_loss(lambda: penalty * tf.reduce_sum(tf.square(layer.trainable_variables[0])

有关更多详细信息，请查看官方文档 Here