在培训期间，我遇到了以下错误。”ValueError:没有为任何变量提供梯度“

Question

我计算了梯度，但我仍然没有得到梯度误差，即使在看了类似错误的答案之后，我也无法找出我遗漏的地方。 我用的是张量流2。 到我的代码的链接是https://github.com/Gadamsetty/ML_practise/blob/master/translatron_test.py

误差如下

ValueError: in converted code:

    <ipython-input-15-031ef53603dd>:92 train_step  *
        optimizer.apply_gradients(zip(gradients,variables))
    /tensorflow-2.0.0/python3.6/tensorflow_core/python/keras/optimizer_v2/optimizer_v2.py:427 apply_gradients
        grads_and_vars = _filter_grads(grads_and_vars)
    /tensorflow-2.0.0/python3.6/tensorflow_core/python/keras/optimizer_v2/optimizer_v2.py:1025 _filter_grads
        ([v.name for _, v in grads_and_vars],))

    ValueError: No gradients provided for any variable: ['encoder_rnn_3/layer0/forward_gru_12/kernel:0', 'encoder_rnn_3/layer0/forward_gru_12/recurrent_kernel:0', 'encoder_rnn_3/layer0/forward_gru_12/bias:0', 'encoder_rnn_3/layer0/backward_gru_12/kernel:0', 'encoder_rnn_3/layer0/backward_gru_12/recurrent_kernel:0', 'encoder_rnn_3/layer0/backward_gru_12/bias:0', 'encoder_rnn_3/layer1/forward_gru_13/kernel:0', 'encoder_rnn_3/layer1/forward_gru_13/recurrent_kernel:0', 'encoder_rnn_3/layer1/forward_gru_13/bias:0', 'encoder_rnn_3/layer1/backward_gru_13/kernel:0', 'encoder_rnn_3/layer1/backward_gru_13/recurrent_kernel:0', 'encoder_rnn_3/layer1/backward_gru_13/bias:0', 'encoder_rnn_3/layer2/forward_gru_14/kernel:0', 'encoder_rnn_3/layer2/forward_gru_14/recurrent_kernel:0', 'encoder_rnn_3/layer2/forward_gru_14/bias:0', 'encoder_rnn_3/layer2/backward_gru_14/kernel:0', 'encoder_rnn_3/layer2/backward_gru_14/recurrent_kernel:0', 'encoder_rnn_3/layer2/backward_gru_14/bias:0', 'encoder_rnn_3/layer3/forward_gru_15/kernel:0', 'encoder_rnn_3/layer3/forward_gru_15/recurrent_kernel:0', 'encoder_rnn_3/layer3/forward_gru_15/bias:0', 'encoder_rnn_3/layer3/backward_gru_15/kernel:0', 'encoder_rnn_3/layer3/backward_gru_15/recurrent_kernel:0', 'encoder_rnn_3/layer3/backward_gru_15/bias:0', 'decoder_rnn_3/bidirectional_6/forward_decoder_layer1/kernel:0', 'decoder_rnn_3/bidirectional_6/forward_decoder_layer1/recurrent_kernel:0', 'decoder_rnn_3/bidirectional_6/forward_decoder_layer1/bias:0', 'decoder_...

Answer 1

由于没有对渐变进行任何操作，因此可以尝试直接使用optimizer.minimize(loss)，而不是先计算渐变，然后分别应用它们(第159-161行）