Question

使用tf.keras.layers.BatchNormalization（）给出 TypeError：不兼容的类型：vs. int64。值为0

使用此链接进行错误再现：： https://colab.research.google.com/drive/16IEd1GjgvPIiG0xnB3ado11u4jLHQZLY

我想使其易于训练。

  enc = tf.keras.layers.TimeDistributed(tf.keras.layers.Conv2D(32, kernel_size=3, activation='relu',padding = 'same',kernel_initializer='glorot_uniform'))(enc)
  print('encoder_mask', enc.shape)

  enc = tf.keras.layers.TimeDistributed(tf.keras.layers.BatchNormalization(epsilon=BN_EPSILON, momentum=BN_MIMENTUM))(enc)
  print('encoder_mask', enc.shape)

  enc = tf.keras.layers.TimeDistributed(tf.keras.layers.Conv2D(16, activation='relu',kernel_size=3,kernel_initializer='glorot_uniform'))(enc)
  print('encoder_mask', enc.shape)

编译后，即使不使用这些变量（BN_EPSILON，BN_MIMENTUM），也会出现以下错误。

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-13-c9408a385d78> in <module>()
      1 lr = 0.0001
----> 2 train_op = tf.train.AdamOptimizer(lr).minimize(reconstuction_loss)

7 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/training/optimizer.py in minimize(self, loss, global_step, var_list, gate_gradients, aggregation_method, colocate_gradients_with_ops, name, grad_loss)
    401         aggregation_method=aggregation_method,
    402         colocate_gradients_with_ops=colocate_gradients_with_ops,
--> 403         grad_loss=grad_loss)
    404 
    405     vars_with_grad = [v for g, v in grads_and_vars if g is not None]

/usr/local/lib/python3.6/dist-packages/tensorflow/python/training/optimizer.py in compute_gradients(self, loss, var_list, gate_gradients, aggregation_method, colocate_gradients_with_ops, grad_loss)
    510         gate_gradients=(gate_gradients == Optimizer.GATE_OP),
    511         aggregation_method=aggregation_method,
--> 512         colocate_gradients_with_ops=colocate_gradients_with_ops)
    513     if gate_gradients == Optimizer.GATE_GRAPH:
    514       grads = control_flow_ops.tuple(grads)

/usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/gradients_impl.py in gradients(ys, xs, grad_ys, name, colocate_gradients_with_ops, gate_gradients, aggregation_method, stop_gradients, unconnected_gradients)
    156         ys, xs, grad_ys, name, colocate_gradients_with_ops,
    157         gate_gradients, aggregation_method, stop_gradients,
--> 158         unconnected_gradients)
    159   # pylint: enable=protected-access
    160 

/usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/gradients_util.py in _GradientsHelper(ys, xs, grad_ys, name, colocate_gradients_with_ops, gate_gradients, aggregation_method, stop_gradients, unconnected_gradients, src_graph)
    718               # issue here because of zeros.
    719               if loop_state:
--> 720                 out_grads[i] = loop_state.ZerosLike(op, i)
    721               else:
    722                 out_grads[i] = control_flow_ops.ZerosLikeOutsideLoop(op, i)

/usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/control_flow_ops.py in ZerosLike(self, op, index)
   1229       # If the shape is known statically, just create a zero tensor with
   1230       # the right shape in the grad loop context.
-> 1231       result = constant_op.constant(0, shape=shape.dims, dtype=val.dtype)
   1232       if dead_branch:
   1233         # op is a cond switch. Guard the zero tensor with a switch.

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/constant_op.py in constant(value, dtype, shape, name)
    244   """
    245   return _constant_impl(value, dtype, shape, name, verify_shape=False,
--> 246                         allow_broadcast=True)
    247 
    248 

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/constant_op.py in _constant_impl(value, dtype, shape, name, verify_shape, allow_broadcast)
    282       tensor_util.make_tensor_proto(
    283           value, dtype=dtype, shape=shape, verify_shape=verify_shape,
--> 284           allow_broadcast=allow_broadcast))
    285   dtype_value = attr_value_pb2.AttrValue(type=tensor_value.tensor.dtype)
    286   const_tensor = g.create_op(

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/tensor_util.py in make_tensor_proto(values, dtype, shape, verify_shape, allow_broadcast)
    499                             dtype.base_dtype != numpy_dtype.base_dtype):
    500     raise TypeError("Incompatible types: %s vs. %s. Value is %s" %
--> 501                     (dtype, nparray.dtype, values))
    502 
    503   # If shape is not given, get the shape from the numpy array.

TypeError: Incompatible types: <dtype: 'resource'> vs. int64. Value is 0

通过运行以下命令来重现错误：

batch_size = 20
inp = tf.placeholder(tf.float32, [batch_size, 19, 64, 64, 3])
out = tf.placeholder(tf.float32, [batch_size, 19, 60, 60, 16])
def model(inp):

  enc = tf.keras.layers.TimeDistributed(tf.keras.layers.Conv2D(128, activation='relu', kernel_size=3,kernel_initializer='glorot_uniform'))(inp)
  enc = tf.keras.layers.TimeDistributed(tf.keras.layers.BatchNormalization())(enc)
  enc = tf.keras.layers.TimeDistributed(tf.keras.layers.Conv2D(16, activation='relu',kernel_size=3,kernel_initializer='glorot_uniform'))(enc)
  return enc

pred = model(inp)

loss = tf.reduce_mean(tf.keras.backend.binary_crossentropy(out, pred))
lr = 0.0001
train_op = tf.train.AdamOptimizer(lr).minimize(loss)

Answer 1

看看示例的固定版本：

https://colab.research.google.com/drive/19DQlzeD2oN8KU37SUFexRzEwrhQdyAe5

您没有正确使用keras功能API-这是常规的方法。您可以找到更多信息in this tf.keras guide

当我使用tf.keras.layers.BatchNormalization（）时，它将引发类型错误

1 个答案: