Gradients with TensorArray fail after scatter of 0 size

[taerimmkim]

system information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): yes

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 16.04

• TensorFlow installed from (source or binary): binary

Describe the problem

아래에서 오류가 발생:

import tensorflow as tf

x_orig = tf.zeros([3], tf.float32)

ta = tf.TensorArray(tf.float32, size=1, element_shape=[3])
ta = ta.write(0, x_orig)
ta = ta.scatter(tf.range(0, 0), ta.gather(tf.range(0, 0)))
x = ta.read(0)
g, = tf.gradients(tf.reduce_sum(x), [x_orig])

sess = tf.Session()
print sess.run(g)

다음과 같은 에러메세지 발생:

tensorflow.python.framework.errors_impl.UnimplementedError: TensorArray has size zero, but element shape <unknown> is not fully defined. Currently only static shapes are supported when packing zero-size TensorArrays.
         [[Node: gradients/TensorArrayScatter/TensorArrayScatterV3_grad/TensorArrayGatherV3 = TensorArrayGatherV3[dtype=DT_FLOAT, element_shape=<unknown>, _device="/job:localhost/replica:0/task:0/cpu:0"](gradients/TensorArrayScatter/TensorArrayScatterV3_grad/TensorArrayGrad/TensorArrayGradV3, range, gradients/TensorArrayScatter/TensorArrayScatterV3_grad/TensorArrayGrad/gradient_flow)]]

ta = ta.scatter(tf.range(0, 0), ta.gather(tf.range(0, 0))) 코드가 수행되면 ''''ta'''' 가 제거된다.
원인은 _TensorArrayScatterGrad가 .scatter가 호출될 때 TensorArray를 element_size만큼 복제하지 못하기 때문인 것 같다.