这是尝试设置简单的RNN：
import tensorflow as tf
BATCH_SIZE = 7
SEQUENCE_LENGTH = 5
VECTOR_SIZE = 3
STATE_SIZE = 4
x = tf.placeholder(tf.int32, [BATCH_SIZE, SEQUENCE_LENGTH, VECTOR_SIZE],
name='input_placeholder')
y = tf.placeholder(tf.int32, [BATCH_SIZE, SEQUENCE_LENGTH,],
name='labels_placeholder')
init_state = tf.zeros([BATCH_SIZE, STATE_SIZE])
rnn_inputs = tf.unstack(x, axis = 1)
cell = tf.contrib.rnn.BasicLSTMCell(STATE_SIZE, state_is_tuple = True)
rnn_outputs, final_state = tf.contrib.rnn.static_rnn(cell, rnn_inputs, initial_state=init_state)
它会出现以下错误：
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-67-cf3e657dfb23> in <module>()
13 cell = tf.contrib.rnn.BasicLSTMCell(STATE_SIZE, state_is_tuple = True)
14 #rnn_outputs, final_state = tf.contrib.rnn.static_rnn(cell, rnn_inputs, initial_state=init_state)
---> 15 results__ = tf.contrib.rnn.static_rnn(cell, rnn_inputs, initial_state=init_state)
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn.py in static_rnn(cell, inputs, initial_state, dtype, sequence_length, scope)
1235 state_size=cell.state_size)
1236 else:
-> 1237 (output, state) = call_cell()
1238
1239 outputs.append(output)
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn.py in <lambda>()
1222 varscope.reuse_variables()
1223 # pylint: disable=cell-var-from-loop
-> 1224 call_cell = lambda: cell(input_, state)
1225 # pylint: enable=cell-var-from-loop
1226 if sequence_length is not None:
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn_cell_impl.py in __call__(self, inputs, state, scope)
178 with vs.variable_scope(vs.get_variable_scope(),
179 custom_getter=self._rnn_get_variable):
--> 180 return super(RNNCell, self).__call__(inputs, state)
181
182 def _rnn_get_variable(self, getter, *args, **kwargs):
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\layers\base.py in __call__(self, inputs, *args, **kwargs)
448 # Check input assumptions set after layer building, e.g. input shape.
449 self._assert_input_compatibility(inputs)
--> 450 outputs = self.call(inputs, *args, **kwargs)
451
452 # Apply activity regularization.
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn_cell_impl.py in call(self, inputs, state)
395 # Parameters of gates are concatenated into one multiply for efficiency.
396 if self._state_is_tuple:
--> 397 c, h = state
398 else:
399 c, h = array_ops.split(value=state, num_or_size_splits=2, axis=1)
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\framework\ops.py in __iter__(self)
474 TypeError: when invoked.
475 """
--> 476 raise TypeError("'Tensor' object is not iterable.")
477
478 def __bool__(self):
TypeError: 'Tensor' object is not iterable.
我不确定为什么会产生此错误。在SO上有这样的多个帖子，但他们没有回答这个问题。
更新
这是我在亚伦回答之后尝试的：
BATCH_SIZE = 7
SEQUENCE_LENGTH = 5
VECTOR_SIZE = 3
STATE_SIZE = 4
x = tf.placeholder(tf.int32, [BATCH_SIZE, SEQUENCE_LENGTH, VECTOR_SIZE],
name='input_placeholder')
y = tf.placeholder(tf.int32, [BATCH_SIZE, SEQUENCE_LENGTH,],
name='labels_placeholder')
init_state_state = tf.zeros([BATCH_SIZE, STATE_SIZE], tf.int32)
init_state_input = tf.zeros([BATCH_SIZE, VECTOR_SIZE], tf.int32)
rnn_inputs = tf.unstack(x, axis = 1)
cell = tf.contrib.rnn.BasicLSTMCell(STATE_SIZE, state_is_tuple = True)
rnn_outputs, final_state = tf.contrib.rnn.static_rnn(cell, rnn_inputs, initial_state=(init_state_state, init_state_input))
现在我有以下错误：
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-2-79fd4965953c> in <module>()
14 rnn_inputs = tf.unstack(x, axis = 1)
15 cell = tf.contrib.rnn.BasicLSTMCell(STATE_SIZE, state_is_tuple = True)
---> 16 rnn_outputs, final_state = tf.contrib.rnn.static_rnn(cell, rnn_inputs, initial_state=(init_state_state, init_state_input))
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn.py in static_rnn(cell, inputs, initial_state, dtype, sequence_length, scope)
1235 state_size=cell.state_size)
1236 else:
-> 1237 (output, state) = call_cell()
1238
1239 outputs.append(output)
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn.py in <lambda>()
1222 varscope.reuse_variables()
1223 # pylint: disable=cell-var-from-loop
-> 1224 call_cell = lambda: cell(input_, state)
1225 # pylint: enable=cell-var-from-loop
1226 if sequence_length is not None:
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn_cell_impl.py in __call__(self, inputs, state, scope)
178 with vs.variable_scope(vs.get_variable_scope(),
179 custom_getter=self._rnn_get_variable):
--> 180 return super(RNNCell, self).__call__(inputs, state)
181
182 def _rnn_get_variable(self, getter, *args, **kwargs):
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\layers\base.py in __call__(self, inputs, *args, **kwargs)
448 # Check input assumptions set after layer building, e.g. input shape.
449 self._assert_input_compatibility(inputs)
--> 450 outputs = self.call(inputs, *args, **kwargs)
451
452 # Apply activity regularization.
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn_cell_impl.py in call(self, inputs, state)
399 c, h = array_ops.split(value=state, num_or_size_splits=2, axis=1)
400
--> 401 concat = _linear([inputs, h], 4 * self._num_units, True)
402
403 # i = input_gate, j = new_input, f = forget_gate, o = output_gate
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn_cell_impl.py in _linear(args, output_size, bias, bias_initializer, kernel_initializer)
1051 _BIAS_VARIABLE_NAME, [output_size],
1052 dtype=dtype,
-> 1053 initializer=bias_initializer)
1054 return nn_ops.bias_add(res, biases)
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\variable_scope.py in get_variable(name, shape, dtype, initializer, regularizer, trainable, collections, caching_device, partitioner, validate_shape, use_resource, custom_getter)
1063 collections=collections, caching_device=caching_device,
1064 partitioner=partitioner, validate_shape=validate_shape,
-> 1065 use_resource=use_resource, custom_getter=custom_getter)
1066 get_variable_or_local_docstring = (
1067 """%s
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\variable_scope.py in get_variable(self, var_store, name, shape, dtype, initializer, regularizer, reuse, trainable, collections, caching_device, partitioner, validate_shape, use_resource, custom_getter)
960 collections=collections, caching_device=caching_device,
961 partitioner=partitioner, validate_shape=validate_shape,
--> 962 use_resource=use_resource, custom_getter=custom_getter)
963
964 def _get_partitioned_variable(self,
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\variable_scope.py in get_variable(self, name, shape, dtype, initializer, regularizer, reuse, trainable, collections, caching_device, partitioner, validate_shape, use_resource, custom_getter)
358 reuse=reuse, trainable=trainable, collections=collections,
359 caching_device=caching_device, partitioner=partitioner,
--> 360 validate_shape=validate_shape, use_resource=use_resource)
361 else:
362 return _true_getter(
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\rnn_cell_impl.py in _rnn_get_variable(self, getter, *args, **kwargs)
181
182 def _rnn_get_variable(self, getter, *args, **kwargs):
--> 183 variable = getter(*args, **kwargs)
184 trainable = (variable in tf_variables.trainable_variables() or
185 (isinstance(variable, tf_variables.PartitionedVariable) and
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\variable_scope.py in _true_getter(name, shape, dtype, initializer, regularizer, reuse, trainable, collections, caching_device, partitioner, validate_shape, use_resource)
350 trainable=trainable, collections=collections,
351 caching_device=caching_device, validate_shape=validate_shape,
--> 352 use_resource=use_resource)
353
354 if custom_getter is not None:
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\variable_scope.py in _get_single_variable(self, name, shape, dtype, initializer, regularizer, partition_info, reuse, trainable, collections, caching_device, validate_shape, use_resource)
723 caching_device=caching_device,
724 dtype=variable_dtype,
--> 725 validate_shape=validate_shape)
726 self._vars[name] = v
727 logging.vlog(1, "Created variable %s with shape %s and init %s", v.name,
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\variables.py in __init__(self, initial_value, trainable, collections, validate_shape, caching_device, name, variable_def, dtype, expected_shape, import_scope)
197 name=name,
198 dtype=dtype,
--> 199 expected_shape=expected_shape)
200
201 def __repr__(self):
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\variables.py in _init_from_args(self, initial_value, trainable, collections, validate_shape, caching_device, name, dtype, expected_shape)
275 with ops.name_scope("Initializer"), ops.device(None):
276 self._initial_value = ops.convert_to_tensor(
--> 277 initial_value(), name="initial_value", dtype=dtype)
278 shape = (self._initial_value.get_shape()
279 if validate_shape else tensor_shape.unknown_shape())
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\variable_scope.py in <lambda>()
699 initializer = initializer(dtype=dtype)
700 init_val = lambda: initializer( # pylint: disable=g-long-lambda
--> 701 shape.as_list(), dtype=dtype, partition_info=partition_info)
702 variable_dtype = dtype.base_dtype
703
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\ops\init_ops.py in __call__(self, shape, dtype, partition_info, verify_shape)
201 verify_shape = self._verify_shape
202 return constant_op.constant(self.value, dtype=dtype, shape=shape,
--> 203 verify_shape=verify_shape)
204
205 def get_config(self):
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\framework\constant_op.py in constant(value, dtype, shape, name, verify_shape)
100 tensor_value = attr_value_pb2.AttrValue()
101 tensor_value.tensor.CopyFrom(
--> 102 tensor_util.make_tensor_proto(value, dtype=dtype, shape=shape, verify_shape=verify_shape))
103 dtype_value = attr_value_pb2.AttrValue(type=tensor_value.tensor.dtype)
104 const_tensor = g.create_op(
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\framework\tensor_util.py in make_tensor_proto(values, dtype, shape, verify_shape)
374 nparray = np.empty(shape, dtype=np_dt)
375 else:
--> 376 _AssertCompatible(values, dtype)
377 nparray = np.array(values, dtype=np_dt)
378 # check to them.
~\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\tensorflow\python\framework\tensor_util.py in _AssertCompatible(values, dtype)
300 else:
301 raise TypeError("Expected %s, got %s of type '%s' instead." %
--> 302 (dtype.name, repr(mismatch), type(mismatch).__name__))
303
304
TypeError: Expected int32, got 0.0 of type 'float' instead.
很抱歉有另一个更新，但在切换到float32后，我收到另一个错误：
BATCH_SIZE = 7
SEQUENCE_LENGTH = 5
VECTOR_SIZE = 3
STATE_SIZE = 4
x = tf.placeholder(tf.float32, [BATCH_SIZE, SEQUENCE_LENGTH, VECTOR_SIZE],
name='input_placeholder')
y = tf.placeholder(tf.float32, [BATCH_SIZE, SEQUENCE_LENGTH,],
name='labels_placeholder')
init_state_state = tf.zeros([BATCH_SIZE, STATE_SIZE], tf.float32)
init_state_input = tf.zeros([BATCH_SIZE, VECTOR_SIZE], tf.float32)
rnn_inputs = tf.unstack(x, axis = 1)
cell = tf.contrib.rnn.BasicLSTMCell(STATE_SIZE, state_is_tuple = True)
rnn_outputs, final_state = tf.contrib.rnn.static_rnn(cell, rnn_inputs, initial_state=(init_state_state, init_state_input))
以下是错误消息的最后一行：
ValueError: Trying to share variable rnn/basic_lstm_cell/kernel, but specified shape (7, 16) and found shape (6, 16).
答案 0 :(得分：2)
这是因为你的LSTM初始状态应该是一个元组，而你只给它一个零向量。使init状态成为两个零向量的元组，你应该没问题。
在简单的RNN或GRU中，只有一个状态向量。 LSTM在该州有两个向量。