what does the question mark in tensorflow shape mean?

Question

I'm looking into magenta code, and printing its tensor object. I got this result:

Tensor("fully_connected/BiasAdd:0", shape=(?, 38), dtype=float32)

What does this question mark in shape mean?

Could you please tell me how to set a placeholder for such kind of tensor? I mean the shape of the placeholder. Thank you. — ytutow, Aug 27 '17 at 19:15
@ytutow [use None as a dimension when creating the tensor](https://www.tensorflow.org/api_docs/python/tf/TensorShape) — user3125280, Feb 15 '18 at 10:55

score 28 · Accepted Answer · answered Dec 03 '16 at 22:24

28

It means that first dimension is not fixed in the graph and it can vary between run calls

answered Dec 03 '16 at 22:24

Yaroslav Bulatov

Thank you very much. It makes sense because I'm dealing with RNN and sequential data. – Kingston Chan Dec 04 '16 at 00:37
2

Thank you. Could you please tell me how to set a placeholder for such kind of tensor? I mean the shape. – ytutow Aug 27 '17 at 18:59
@yutow for example: img_holder = tf.placeholder(tf.float32, [None, 784]) – user2757572 Apr 03 '19 at 17:30
Can you please in short explain the meaning of "graph" and "calls" regarding your answer? – KCK Jul 06 '19 at 05:57

1 Answers1