len对符号没有很好的定义。请致电`x.shape`而不是`len（x）`获取形状信息

Question

我有一个解码器输出17个维度，其中不同部分是标签和数字。因此，对于标签，我使用了一种热编码，并使用了“ softmax”激活；对于数字，我使用了“ Sigmoid”激活函数。

这里是解码器：

latent_inputs = Input(shape=(latent_dim,), name='z_sampling')
x = Dense(intermediate_dim, activation='relu')(latent_inputs)
x1 = Dense(intermediate_dim, activation='relu')(x)
x2 = Dense(intermediate_dim, activation='relu')(x1)
output1 = Dense(Num_classes, activation='softmax')(x2)
output2 = Dense(3, activation='sigmoid')(x2)
output3 = Dense(Num_classes, activation='softmax')(x2)
output4 = Dense(2, activation='sigmoid')(x2)

outputs = [output1, output2, output3, output4]
# instantiate decoder model
decoder = Model(latent_inputs, outputs, name='decoder')

而且，这是我定义损失函数的方式：

reconstruction_loss = categorical_crossentropy(inputs[0:Num_classes],outputs[0:Num_classes] ) 
+ binary_crossentropy(inputs[Num_classes:Num_classes+3],outputs[Num_classes:Num_classes+3])\
+categorical_crossentropy(inputs[Num_classes+3:2*Num_classes+3],outputs[Num_classes+3:2*Num_classes+3])\
+ binary_crossentropy(inputs[2*Num_classes+3:2*Num_classes+5],outputs[2*Num_classes+3:2*Num_classes+5])

但是，我在这个问题上弄错了。您能帮忙吗？

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-19-4de2050a20eb> in <module>
     17         reconstruction_loss = mse(inputs, outputs)
     18     else:
---> 19         reconstruction_loss = categorical_crossentropy(inputs[0:Num_classes],outputs[0:Num_classes] )
     20         + binary_crossentropy(inputs[Num_classes:Num_classes+3],outputs[Num_classes:Num_classes+3])\
     21         + categorical_crossentropy(inputs[Num_classes+3:2*Num_classes+3],outputs[Num_classes+3:2*Num_classes+3])\

C:\ProgramData\Anaconda3\lib\site-packages\keras\losses.py in categorical_crossentropy(y_true, y_pred, from_logits, label_smoothing)
    678 
    679 def categorical_crossentropy(y_true, y_pred, from_logits=False, label_smoothing=0):
--> 680     y_pred = K.constant(y_pred) if not K.is_tensor(y_pred) else y_pred
    681     y_true = K.cast(y_true, y_pred.dtype)
    682 

C:\ProgramData\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py in constant(value, dtype, shape, name)
    647     with tf_ops.init_scope():
    648         return tf_keras_backend.constant(
--> 649             value, dtype=dtype, shape=shape, name=name)
    650 
    651 

C:\ProgramData\Anaconda3\lib\site-packages\tensorflow_core\python\keras\backend.py in constant(value, dtype, shape, name)
    935     dtype = floatx()
    936 
--> 937   return constant_op.constant(value, dtype=dtype, shape=shape, name=name)
    938 
    939 

C:\ProgramData\Anaconda3\lib\site-packages\tensorflow_core\python\framework\constant_op.py in constant(value, dtype, shape, name)
    256   """
    257   return _constant_impl(value, dtype, shape, name, verify_shape=False,
--> 258                         allow_broadcast=True)
    259 
    260 

C:\ProgramData\Anaconda3\lib\site-packages\tensorflow_core\python\framework\constant_op.py in _constant_impl(value, dtype, shape, name, verify_shape, allow_broadcast)
    264   ctx = context.context()
    265   if ctx.executing_eagerly():
--> 266     t = convert_to_eager_tensor(value, ctx, dtype)
    267     if shape is None:
    268       return t

C:\ProgramData\Anaconda3\lib\site-packages\tensorflow_core\python\framework\constant_op.py in convert_to_eager_tensor(value, ctx, dtype)
     94       dtype = dtypes.as_dtype(dtype).as_datatype_enum
     95   ctx.ensure_initialized()
---> 96   return ops.EagerTensor(value, ctx.device_name, dtype)
     97 
     98 

ValueError: TypeError: len is not well defined for symbolic Tensors. (decoder/dense_7/Softmax:0) Please call `x.shape` rather than `len(x)` for shape information.
Traceback (most recent call last):

  File "C:\ProgramData\Anaconda3\lib\site-packages\tensorflow_core\python\framework\ops.py", line 733, in __len__
    "shape information.".format(self.name))

TypeError: len is not well defined for symbolic Tensors. (decoder/dense_7/Softmax:0) Please call `x.shape` rather than `len(x)` for shape information.

Answer 1

为了解决这个问题，我为编码器定义了四个输入，如下所示：

input1 = Input(shape=(Num_classes,), name='encoder_input1')
input2 = Input(shape=(3,), name='encoder_input2')
input3 = Input(shape=(Num_classes,), name='encoder_input3')
input4 = Input(shape=(2,), name='encoder_input4')
added = keras.layers.concatenate([input1, input2, input3, input4], axis = -1)
x = Dense(128, activation='relu')(added)
.
.
.

我用四个输出定义了解码器：

.
.
.
x3 = Dense(128, activation='relu')(x2)
out1 = Dense(Num_classes, activation='softmax',name='label1')(x3)
out2 = Dense(3, activation='sigmoid',name='dim1')(x3)
out3 = Dense(Num_classes, activation='softmax',name='label2')(x3)
out4 = Dense(2, activation='sigmoid',name='dim2')(x3)

以及损失函数如下：

reconstruction_loss = categorical_crossentropy(input1,out1) +\ 
binary_crossentropy(input2,out2)+\
categorical_crossentropy(input3,out3) +\
binary_crossentropy(input4,out4)