我想编写脚本来构建Tensorflow图，该图可以与不同大小的tf.placeholder一起使用。我运行以下脚本时得到InvalidArgumentError：
# Loading training and validation data
# train data are in x_train, y_train
# validation data are in x_valid, y_valid
...
with tf.variable_scope("model", reuse=None):
# Build a Graph that computes the logits
network = Model(
batch_size=BATCH_SIZE,
input_dims=INPUT_DIMS,
seq_length=SEQ_LENGTH,
num_classes=NUM_CLASSES !!!!! ----> line 528, in main_seq
)
# Calculate loss
network.loss_op = softmax_seq_loss_by_example(
logits=network.model,
labels=network.targets,
batch_size=network.batch_size,
seq_length=network.seq_length
)
# Calculate predictions
network.pred_op = tf.argmax(network.model, 1)
with tf.variable_scope("model", reuse=True):
# Build a Graph that computes the logits
network_valid = Model(
batch_size=1,
input_dims=INPUT_DIMS,
seq_length=1,
num_classes=NUM_CLASSES
)
# Calculate loss
network_valid.loss_op = softmax_seq_loss_by_example(
logits=network_valid.model,
labels=network_valid.targets,
batch_size=network_valid.batch_size,
seq_length=network_valid.seq_length
)
# Calculate predictions
network_valid.pred_op = tf.argmax(network_valid.model, 1)
with tf.Session() as sess:
# Initialize variables in the graph
sess.run(tf.initialize_all_variables())
for epoch in xrange(n_epochs):
# Update parameters and compute loss of training set
y_true_train, y_pred_train, train_loss, train_duration = \
run_seq_epoch(
sess=sess, network=network,
inputs=x_train, targets=y_train,
train_op=train_op,
is_train=True
)
# Evaluate the model on the validation set
y_true_val, y_pred_val, valid_loss, valid_duration = \
run_seq_epoch(
sess=sess, network=network_valid, **** ----> replace_location
inputs=x_valid, targets=x_valid,
train_op=tf.no_op(),
is_train=False !!!!! ----> line 644, in main_seq
)
...
功能run_seq_epoch和课程Model如下所示：
class Model(object):
def __init__(self, batch_size, input_dims, seq_length, num_classes):
self.batch_size = batch_size
self.input_dims = input_dims
self.seq_length = seq_length
self.num_classes = num_classes
# Operations to compute loss and prediction, which will be assigned
# after the initialization
self.loss_op = None
self.pred_op = None
# Placeholder for input data
self.inputs = tf.placeholder(
tf.float32,
shape=[batch_size*seq_length, input_dims, 1, 1]
)
self.targets = tf.placeholder(
tf.int32,
shape=[batch_size*seq_length, ]
)
self.is_train = tf.placeholder(tf.bool)
# Use the defined placeholder above create a model
...
def run_seq_epoch(sess, network, inputs, targets, train_op, is_train):
...
# Initial state for LSTM
state = network.initial_state.eval()
for x_batch, y_batch in iterate_batch_seq_minibatches(inputs,
targets,
network.batch_size,
network.seq_length):
feed_dict = {
network.inputs: x_batch,
network.targets: y_batch,
network.initial_state: state,
network.is_train: is_train
}
_, loss_value, y_pred, state = sess.run(
[train_op, network.loss_op, network.pred_op, network.final_state],
feed_dict=feed_dict !!!!! ----> line 471, in run_seq_epoch
)
...
return _y_true, y_pred, loss, duration
以下是network和network_valid中占位符的名称和形状，以及x_batch和y_batch的形状和类型，用于培训和验证：
# Placeholder of network.inputs and network.targets
model/Placeholder:0: (200, 7680, 1, 1)
model/Placeholder_1:0: (200,)
# x_batch and y_batch for training
x_batch.shape: (200, 7860, 1, 1), dtype=float32
y_batch.shape: (200, ), dtype=int32
# Placeholder of network_valid.inputs and network_valid.targets
model_1/Placeholder:0: (1, 7680, 1, 1)
model_1/Placeholder_1:0: (1,)
# x_batch and y_batch for validation
x_batch.shape: (1, 7860, 1, 1), type=float32
y_batch.shape: (1, ), type=int32
但是，当我运行脚本时，我为tensorflow.python.framework.errors.InvalidArgumentError占位符获得了targets：
Traceback (most recent call last):
File "main.py", line 789, in <module>
fold_idx=fold_idx
File "main.py", line 644, in main_seq
is_train=False
File "main.py", line 471, in run_seq_epoch
feed_dict=feed_dict
File "/home/akara/miniconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 372, in run
run_metadata_ptr)
File "/home/akara/miniconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 636, in _run
feed_dict_string, options, run_metadata)
File "/home/akara/miniconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 708, in _do_run
target_list, options, run_metadata)
File "/home/akara/miniconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 728, in _do_call
raise type(e)(node_def, op, message)
tensorflow.python.framework.errors.InvalidArgumentError: You must feed a value for placeholder tensor 'model/Placeholder_1' with dtype int32 and shape [200]
[[Node: model/Placeholder_1 = Placeholder[dtype=DT_INT32, shape=[200], _device="/job:localhost/replica:0/task:0/gpu:0"]()]]
[[Node: model_1/total_loss/_329 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:localhost/replica:0/task:0/gpu:0", send_device_incarnation=1, tensor_name="edge_1326_model_1/total_loss", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"]()]]
Caused by op u'model/Placeholder_1', defined at:
File "main.py", line 789, in <module>
fold_idx=fold_idx
File "main.py", line 528, in main_seq
num_classes=NUM_CLASSES
File "/home/akara/Workspace/project/model.py", line 62, in __init__
shape=[batch_size*seq_length, ]
File "/home/akara/miniconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/array_ops.py", line 895, in placeholder
name=name)
File "/home/akara/miniconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/gen_array_ops.py", line 1238, in _placeholder
name=name)
File "/home/akara/miniconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/op_def_library.py", line 704, in apply_op
op_def=op_def)
File "/home/akara/miniconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 2260, in create_op
original_op=self._default_original_op, op_def=op_def)
File "/home/akara/miniconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1230, in __init__
self._traceback = _extract_stack()
如果我多次重新运行它，它会报告相同的错误，但使用不同的占位符：
is_train占位符
...
tensorflow.python.framework.errors.InvalidArgumentError: You must feed a value for placeholder tensor 'model/Placeholder_2' with dtype bool
[[Node: model/Placeholder_2 = Placeholder[dtype=DT_BOOL, shape=[], _device="/job:localhost/replica:0/task:0/gpu:0"]()]]
[[Node: _recv_model_1/Placeholder_0/_349 = _Send[T=DT_FLOAT, client_terminated=false, recv_device="/job:localhost/replica:0/task:0/gpu:0", send_device="/job:localhost/replica:0/task:0/cpu:0", send_device_incarnation=1, tensor_name="edge_2567__recv_model_1/Placeholder_0", _device="/job:localhost/replica:0/task:0/cpu:0"](_recv_model_1/Placeholder_0)]]
Caused by op u'model/Placeholder_2', defined at:
...
inputs占位符
...
tensorflow.python.framework.errors.InvalidArgumentError: You must feed a value for placeholder tensor 'model/Placeholder' with dtype float and shape [200,7680,1,1]
[[Node: model/Placeholder = Placeholder[dtype=DT_FLOAT, shape=[200,7680,1,1], _device="/job:localhost/replica:0/task:0/gpu:0"]()]]
[[Node: model_1/ArgMax/_354 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:localhost/replica:0/task:0/gpu:0", send_device_incarnation=1, tensor_name="edge_105_model_1/ArgMax", tensor_type=DT_INT64, _device="/job:localhost/replica:0/task:0/cpu:0"]()]]
Caused by op u'model/Placeholder', defined at:
...
如果我将network_valid替换为 replace_location 中的network，则这些脚本完美运作。因此，当我使用network_valid时，我不确定为什么会出现错误。
更新
我还发现错误来自model变量范围的占位符（即network用于培训），而不是model_1范围（即network_valid验证），即使脚本在验证期间停止运行。
我重命名了占位符并将它们移到Model类之外。然后我在Tensorboard（Graph from Tensorboard）中查看我的图表。似乎network占位符与network_valid之间没有链接，类似于network_valid占位符。
