Error evaluating a TensorArray in a while loop

Question

I've built the following TensorArray:

ta = tf.TensorArray(
    dtype=tf.float32,   
    size=0,
    dynamic_size=True,
    element_shape=tf.TensorShape([None, None])
)

and called ta = ta.write(idx, my_tensor) inside a while_loop.

When evaluating the output = ta.stack() tensor in a session, I receive this error message:

ValueError: Cannot use '.../TensorArrayWrite/TensorArrayWriteV3' as input to '.../TensorArrayStack_1/TensorArraySizeV3' because '.../TensorArrayWrite/TensorArrayWriteV3' is in a while loop. See info log for more details.

I don't understand this error message, could you please help me ?

Update: A minimal example might be difficult to come up with, but this is what I am doing: I am using the reference to the ta TensorArray inside the cell_input_fn of AttentionWrapper. This callback is used in AttentionWrapper's call method, where another TensorArray named alignment_history is being written. Therefore the while_loop code is not designed by me, it's part of the TF dynamic RNN computation tf.nn.dynamic_rnn.

Hard to tell without any more code. Can you please post a full minimal example of the issue? — javidcf
– javidcf, Commented Feb 11, 2019 at 14:08
@jdehesa I updated my post a few hours ago, please let me now if this is clear enough. — George
– George, Commented Feb 11, 2019 at 21:06

Eugene Brevdo · Accepted Answer · 2019-02-11 22:34:00Z

1

Not sure if this is what's biting you, but you have to make sure your while_loop function takes the tensor array as input and emits an updated one as output; and you have to use the final version of the TensorArray at the end of the while_loop:

def fn(ta_old):
  return ta_old.write(...)

ta_final = while_loop(..., body=fn, [tf.TensorArray(...)])

values = ta_final.stack()

specifically you should never access ta_old outside of fn().

answered Feb 11, 2019 at 22:34

Eugene Brevdo

8997 silver badges8 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

George Over a year ago

The error gets triggered by the size call in _GraphTensorArray, which comes from stack. Since you know well the AttentionWrapper code, would it be possible to make cell_input_fn use the same RNN while_loop and update a TensorArray at the same time with the alignment_history TensorArray ?

George Over a year ago

Based on your comment, I think that I understand: the while_loop inside dynamic_rnn takes the cell state, an AttentionWrapperState, as one of the loop variables. Hence my TensorArray referenced in cell_input_fn does not get added to the loop vars. Would need a better design.

Collectives™ on Stack Overflow

Error evaluating a TensorArray in a while loop

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related