TensorFlow has no attribute “with_dependencies”

Question

I want to use the tf.with_dependencies function to save the state of my RNNs. For some reason I get the following error.

Answer 1

Or try: from tensorflow.python.ops.control_flow_ops import with_dependencies

Answer 2

There is no such function in the TensorFlow API. Instead you can use with tf.control_dependencies(): and tf.identity() to achieve the intended effect:

with tf.control_dependencies([expected_output]):
  result = tf.identity(input_tensor)

Answer 3

tf.with_dependencies was deprecated somewhere in the end of 2015. Nonetheless it is still defined in the tf code, it is no longer exported (no @tf_export in front of the function) and therefore is not available.

Use

with tf.control_dependencies([expected_output]):
  result = tf.identity(input_tensor)

as suggested by mrry, as it does absolutely the same thing.