Question

我有一个模型，该模型需要两个输入（分别为[50x2特征]和连续[50x714特征]变量）。我想使用以下代码来计算相对于目标的输入梯度：

features_tens = [tf.convert_to_tensor(x) for x in features]
def get_gradients(data_input):
    with tf.GradientTape() as tape:
        tape.watch(data_input)
        # Run the forward pass of the layer.
        # The operations that the layer applies
        # to its inputs are going to be recorded
        # on the GradientTape.
        preds = tf_model(data_input, training=False)
    
    # Use the gradient tape to automatically retrieve
    # the gradients of the trainable variables with respect to the target.    
    grads = tape.gradient(target=preds, sources=data_input)
return grads

grads = get_gradients(data_input=features_tens)

应届毕业生的毕业证书为：

[None, <tf.Tensor: id=1237340, shape=(50, 714), dtype=float32, numpy=
array([[ 3.2849261e-05,  1.1156656e+00, -1.0219267e-02, ...,
    -6.4205074e-01, -1.5517103e-02, -5.7817552e-02],
   [ 1.6995892e-04,  1.0363499e+00, -8.8197924e-02, ...,
    -5.7580328e-01,  6.5039340e-03,  1.1646263e-01],
   [-2.5057874e-04,  1.4367001e+00, -1.3081819e-02, ...,
    -6.8868393e-01, -3.5415408e-01,  2.4788140e-01],
   ...,
   [ 2.5332673e-04,  1.0083977e+00, -3.3813875e-02, ...,
    -4.5079905e-01,  9.5710188e-02,  5.0962836e-01],
   [ 2.3785071e-04,  1.3727474e+00, -4.5678988e-02, ...,
    -6.2601656e-01, -2.9648536e-01,  3.2699940e-01],
   [ 1.5690899e-04,  1.0958921e+00, -1.1370061e-02, ...,
    -3.3662343e-01, -1.6279417e-01,  9.6774891e-02]], dtype=float32)>]

第一部分是None，第二部分是一些渐变，但我不确定它们是否正确。

这到底是怎么回事？

GradientTape（）和具有两个输入的模型对第一个输入返回None，对第二个输入返回东西

0 个答案: