如何将`None` Batch Size传递给tensorflow动态rnn？

Question

我正在尝试构建一个CNN+LSTM+CTC 用于单词识别的模型。
最初我有一个图像，我正在转换使用CNN 在文字图像上提取的特征，并构建一系列特征，我将其作为顺序数据传递给RNN。

以下是我将特征转换为顺序数据的方式：
[[a1,b1,c1],[a2,b2,c2],[a3,b3,c3]] -> [[a1,a2,a3],[b1,b2,b3],[c1,c2,c3]]
其中a,b,c 是使用CNN 提取的3 个特征。
目前我可以将常量batch_size 传递给模型common.BATCH_SIZE，但我想要的是能够将变量batch_size 传递给模型。
这是怎么做到的？

inputs = tf.placeholder(tf.float32, [common.BATCH_SIZE, common.OUTPUT_SHAPE[1], common.OUTPUT_SHAPE[0], 1])
# Here we use sparse_placeholder that will generate a
# SparseTensor required by ctc_loss op.
targets = tf.sparse_placeholder(tf.int32)

# 1d array of size [batch_size]
seq_len = tf.placeholder(tf.int32, [common.BATCH_SIZE])

model = tf.layers.conv2d(inputs, 64, (3,3),strides=(1, 1), padding='same', name='c1')
model = tf.layers.max_pooling2d(model, (3,3), strides=(2,2), padding='same', name='m1')
model = tf.layers.conv2d(model, 128,(3,3), strides=(1, 1), padding='same', name='c2')
model = tf.layers.max_pooling2d(model, (3,3),strides=(2,2), padding='same', name='m2')
model = tf.transpose(model, [3,0,1,2])
shape = model.get_shape().as_list()
model = tf.reshape(model, [shape[0],-1,shape[2]*shape[3]])

cell = tf.nn.rnn_cell.LSTMCell(common.num_hidden, state_is_tuple=True)
cell = tf.nn.rnn_cell.DropoutWrapper(cell, input_keep_prob=0.5, output_keep_prob=0.5)
stack = tf.nn.rnn_cell.MultiRNNCell([cell]*common.num_layers, state_is_tuple=True)

outputs, _ = tf.nn.dynamic_rnn(cell, model, seq_len, dtype=tf.float32,time_major=True)

更新：

batch_size = tf.placeholder(tf.int32, None, name='batch_size')

inputs = tf.placeholder(tf.float32, [batch_size, common.OUTPUT_SHAPE[1], common.OUTPUT_SHAPE[0], 1])
# Here we use sparse_placeholder that will generate a
# SparseTensor required by ctc_loss op.
targets = tf.sparse_placeholder(tf.int32)

# 1d array of size [batch_size]
seq_len = tf.placeholder(tf.int32, [batch_size])

model = tf.layers.conv2d(inputs, 64, (3,3),strides=(1, 1), padding='same', name='c1')
model = tf.layers.max_pooling2d(model, (3,3), strides=(2,2), padding='same', name='m1')
model = tf.layers.conv2d(model, 128,(3,3), strides=(1, 1), padding='same', name='c2')
model = tf.layers.max_pooling2d(model, (3,3),strides=(2,2), padding='same', name='m2')
model = tf.transpose(model, [3,0,1,2])
shape = model.get_shape().as_list()
model = tf.reshape(model, [shape[0],-1,shape[2]*shape[3]])

cell = tf.nn.rnn_cell.LSTMCell(common.num_hidden, state_is_tuple=True)
cell = tf.nn.rnn_cell.DropoutWrapper(cell, input_keep_prob=0.5, output_keep_prob=0.5)
stack = tf.nn.rnn_cell.MultiRNNCell([cell]*common.num_layers, state_is_tuple=True)

outputs, _ = tf.nn.dynamic_rnn(cell, model, seq_len, dtype=tf.float32,time_major=True)

我收到如下错误：

    Traceback (most recent call last):
  File "lstm_and_ctc_ocr_train.py", line 203, in <module>
    train()
  File "lstm_and_ctc_ocr_train.py", line 77, in train
    logits, inputs, targets, seq_len, batch_size = model.get_train_model()
  File "/home/himanshu/learning-tf/tf/code/tensorflow_lstm_ctc_ocr/model.py", line 20, in get_train_model
inputs = tf.placeholder(tf.float32, [batch_size, common.OUTPUT_SHAPE[1], common.OUTPUT_SHAPE[0], 1])
File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/array_ops.py", line 1530, in placeholder
    return gen_array_ops._placeholder(dtype=dtype, shape=shape, name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/gen_array_ops.py", line 1954, in _placeholder
    name=name)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/op_def_library.py", line 705, in apply_op
    attr_value.shape.CopyFrom(_MakeShape(value, key))
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/op_def_library.py", line 198, in _MakeShape
    return tensor_shape.as_shape(v).as_proto()
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/tensor_shape.py", line 798, in as_shape
    return TensorShape(shape)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/tensor_shape.py", line 434, in __init__
    self._dims = [as_dimension(d) for d in dims_iter]
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/tensor_shape.py", line 376, in as_dimension
    return Dimension(value)
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/tensor_shape.py", line 32, in __init__
    self._value = int(value)
TypeError: int() argument must be a string or a number, not 'Tensor'

Answer 1

您应该能够将 batch_size 作为占位符传递给动态 RNN。根据我的经验，您可能会遇到的唯一令人头疼的问题是，如果您没有提前指定它的形状，那么您应该通过 [] 来使事情正常工作，如下所示：

batchsize = tf.placeholder(tf.int32, [], name='batchsize')

然后以通常的方式在 sess.run() 期间输入它的值。这对我来说很有效，同时训练大批量，但随后以 1 的批量生成。

但严格来说，您甚至不需要专门为dynamic_rnn 指定批量大小，对吗？如果您使用 MultiRNNCell 来获得零状态，则需要它，但我没有看到您在代码中这样做...

*** 更新：

正如评论中所讨论的，您的问题似乎与dynamic_rnn 无关，更多的是与您使用占位符inputs 来指定另一个占位符seq_len 的形状有关。这是重现相同错误的代码：

import tensorflow as tf

a = tf.placeholder(tf.int32, None, name='a')
b = tf.placeholder(tf.int32, [a, 5], name='b')
c = b * 5

with tf.Session() as sess:
    C = sess.run(c, feed_dict={a:1, b:[[1,2,3,4,5]]})

这是错误：

TypeError: int() argument must be a string, a bytes-like object or a number, not 'Tensor'

在遇到 dynamic_rnn 的麻烦之前，我建议您找到解决此问题的方法，或者通过更改代码或提出一个单独的问题来了解如何使用占位符来捏造它。