Tensorflow ValueError：形状（？，1）和（？，）不兼容

Question

我在使用 3 个 lstm 层运行我的代码时遇到了这个错误。不知道如何修复它。任何人都可以帮忙。这里 MAX_SEQUENCE_LENGTH=250。运行成本函数后，我收到错误“ValueError: Shapes (?, 1) and (?,) are incompatible'

# Generate a Tensorflow Graph
tf.reset_default_graph()
batch_size = 25
embedding_size = 50
lstmUnits = 64
max_label = 2

x = tf.placeholder(tf.int32, [None, MAX_SEQUENCE_LENGTH])
y = tf.placeholder(tf.int32, [None])

number_of_layers=3

#  Embeddings to represent words
saved_embeddings = np.load('wordVectors.npy')
embeddings = tf.nn.embedding_lookup(saved_embeddings, x)

def lstm_cell():
  return tf.contrib.rnn.BasicLSTMCell(lstmUnits,reuse=tf.get_variable_scope().reuse)

lstmCell = tf.contrib.rnn.MultiRNNCell([lstm_cell() for _ in range(number_of_layers)])

lstmCell = tf.contrib.rnn.DropoutWrapper(cell=lstmCell, output_keep_prob=0.75)

outputs, final_state = tf.nn.dynamic_rnn(lstmCell, embeddings, dtype=tf.float32)

predictions = tf.contrib.layers.fully_connected(outputs[:, -1], 1, activation_fn=tf.sigmoid)

cost = tf.losses.mean_squared_error(y, predictions)

ValueError：形状 (?, 1) 和 (?,) 不兼容 完整的错误信息如下。

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-19-f261b46a6f62> in <module>()
1 # Try 3
----> 2 cost = tf.losses.mean_squared_error(y, predictions)
3 cost
4 #y.shape
5 #y.reshape[]

/home/lavared/anaconda3/lib/python3.5/site-packages/tensorflow/python/ops/losses/losses_impl.py in mean_squared_error(labels, predictions, weights, scope, loss_collection, reduction)
564     predictions = math_ops.to_float(predictions)
565     labels = math_ops.to_float(labels)
--> 566     predictions.get_shape().assert_is_compatible_with(labels.get_shape())
567     losses = math_ops.squared_difference(predictions, labels)
568     return compute_weighted_loss(

/home/lavared/anaconda3/lib/python3.5/site-packages/tensorflow/python/framework/tensor_shape.py in assert_is_compatible_with(self, other)
751     """
752     if not self.is_compatible_with(other):
--> 753       raise ValueError("Shapes %s and %s are incompatible" % (self, other))
754 
755   def most_specific_compatible_shape(self, other):

ValueError: Shapes (?, 1) and (?,) are incompatible

Answer 1

我知道这个问题已经有一个月了。

几天前我遇到了这个问题。这是一个众所周知的bug，尽管他们只解决了那个特定的案例。

在你的情况下，我找到的唯一可行的解​​决方案是修改：

y = tf.placeholder(tf.int32, [None])

在：

y = tf.placeholder(tf.int32, [None, 1])