Tensorflow 梯度是 None （'没有为任何变量提供梯度'）

Question

当我运行以下脚本时，我收到错误消息“没有为任何变量提供渐变”。 “grads”变量是“None”值的列表。这么简单的脚本可能出了什么问题？

import tensorflow as tf
import numpy as np
tf.enable_eager_execution()


class Model(tf.keras.Model):
    def __init__(self):
        super(Model, self).__init__()
        self.layer = tf.keras.layers.Dense(4, activation = "linear")

    def call(self, x):
        y = self.layer(x)

        return y

model = Model()
model._set_inputs(tf.zeros((1, 5)))

optimizer = tf.train.GradientDescentOptimizer(0.5)


# gibberish data
x_train = np.array([[0, 0, 0, 0, 1]], dtype=np.float32)
y_train = np.array([[0.1, 0.1, 0.4, 0.4]])

y_pred = model.call(x_train)

with tf.GradientTape() as tape:
    loss = tf.losses.mean_squared_error(y_train, y_pred)

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

Answer 1

模型预测线

y_pred = model.call(x_train)

必须在 with tf.GradientTape() as tape: 范围内。