Question

我尝试使用以下代码执行急切执行并禁用急切执行：


model = build_encoder_decoder()
final = build_refinement(model)
final.load_weights('/content/gdrive/My Drive/DIM/models/model.01-0.1296.hdf5')
for layer in final.layers:
    layer.trainable = True
import tensorflow as tf
sgd = tensorflow.keras.optimizers.SGD(lr=1e-5, decay=1e-6, momentum=0.9, nesterov=True)
nadam = tensorflow.keras.optimizers.Nadam(lr=2e-5)
tensorflow.compat.v1.disable_eager_execution()
decoder_target = tensorflow.compat.v1.placeholder(dtype='float32', shape=(None, None, None, None))
# final.compile(optimizer=sgd, loss=alpha_prediction_loss)
final.compile(optimizer=sgd,loss=alpha_prediction_loss, target_tensors=[decoder_target],run_eagerly=False)

当我尝试执行它时，出现以下错误：

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-43-912314147d95> in <module>()
----> 1 final.compile(optimizer=sgd,loss=alpha_prediction_loss, target_tensors=[decoder_target],run_eagerly=False)

1 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/engine/training.py in _validate_compile(self, optimizer, metrics, **kwargs)
   2489     if kwargs.pop('target_tensors', None) is not None:
   2490       raise ValueError(
-> 2491           'target_tensors argument is not supported when executing eagerly.')
   2492     invalid_kwargs = set(kwargs) - {
   2493         'experimental_steps_per_execution', 'sample_weight_mode'

ValueError: target_tensors argument is not supported when executing eagerly.

我尝试按照在StackOverflow和其他站点上找到的某些方法禁用急切执行，但我一直收到此错误。有人可以告诉我我要去哪里哪里以及如何纠正这个问题吗？

Answer 1

您正在使用哪个版本的tensorflow？您不需要在2.x版中使用eager_execution

训练模型时，急切执行时不支持获取-值错误target_tensors参数

1 个答案: