Using placeholder as shape in tensorflow

Question

I'm try to define a two dimensional placeholder in tensorflow, However, I don't know the size of that in advance. Hence I define another placeholder, but it seems it doesn't work at all. Here is the minimum example:

import tensorflow as tf

batchSize = tf.placeholder(tf.int32)
input = tf.placeholder(tf.int32, [batchSize, 5])

Error message:

Traceback (most recent call last):
  File "C:/Users/v-zhaom/OneDrive/testconv/test_placeholder.py", line 5, in <module>
    input = tf.placeholder(tf.int32, [batchSize, 5])
  File "C:\Python35\lib\site-packages\tensorflow\python\ops\array_ops.py", line 1579, in placeholder
    shape = tensor_shape.as_shape(shape)
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 821, in as_shape
    return TensorShape(shape)
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 457, in __init__
    self._dims = [as_dimension(d) for d in dims_iter]
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 457, in <listcomp>
    self._dims = [as_dimension(d) for d in dims_iter]
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 378, in as_dimension
    return Dimension(value)
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 33, in __init__
    self._value = int(value)
TypeError: int() argument must be a string, a bytes-like object or a number, not 'Tensor'

Then I tried to pack the shape, so I have this:

    input = tf.placeholder(tf.int32, tf.pack([batchSize, 5]))

doesn't work either:

Traceback (most recent call last):
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 451, in __init__
    dims_iter = iter(dims)
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\ops.py", line 510, in __iter__
    raise TypeError("'Tensor' object is not iterable.")
TypeError: 'Tensor' object is not iterable.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "C:/Users/v-zhaom/OneDrive/testconv/test_placeholder.py", line 5, in <module>
    input = tf.placeholder(tf.int32, tf.pack([batchSize, 5]))
  File "C:\Python35\lib\site-packages\tensorflow\python\ops\array_ops.py", line 1579, in placeholder
    shape = tensor_shape.as_shape(shape)
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 821, in as_shape
    return TensorShape(shape)
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 454, in __init__
    self._dims = [as_dimension(dims)]
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 378, in as_dimension
    return Dimension(value)
  File "C:\Python35\lib\site-packages\tensorflow\python\framework\tensor_shape.py", line 33, in __init__
    self._value = int(value)
TypeError: int() argument must be a string, a bytes-like object or a number, not 'Tensor'

Answer 1

Use None if you don't know the length in some dimension in advance, e.g.

input = tf.placeholder(tf.int32, [None, 5])

When you feed this placeholder a proper array of shape (batch_size, 5), it's dynamic shape will be set correctly, i.e.

sess.run(tf.shape(input), feed_dict={input: np.zeros(dtype=np.int32, shape=(10, 5))})

will return

array([10,  5], dtype=int32)

as expected