tf.tuple

Groups tensors together.

tf.tuple(
    tensors, control_inputs=None, name=None
)

The returned tensors have the same value as the input tensors, but they are computed only after all the input tensors have been computed.

See also tf.group and tf.control_dependencies.

Example:

with tf.Graph().as_default():
  with tf.compat.v1.Session() as sess:
    v = tf.Variable(0.0)
    a = tf.constant(1.0)
    sess.run(tf.compat.v1.global_variables_initializer())
    for i in range(5):
      update_op = v.assign_add(1.0)
      b = a + v
      res_b = sess.run(b)
      res_v = sess.run(v)
      print(res_v)
0.0
0.0
0.0
0.0
0.0
 
with tf.Graph().as_default():
  with tf.compat.v1.Session() as sess:
    v = tf.Variable(0.0)
    a = tf.constant(1.0)
    sess.run(tf.compat.v1.global_variables_initializer())
    for i in range(5):
      update_op = v.assign_add(1.0)
      calc = [a + v]
      # `tf.tuple` ensures `update_op` is run before `b`
      b = tf.tuple(calc, [tf.group(update_op)])
      res_b = sess.run(b)
      res_v = sess.run(v)
      print(res_v)
1.0
2.0
3.0
4.0
5.0

tensors A list of Tensors or IndexedSlices, some entries can be None.
control_inputs List of additional ops to finish before returning.
name (optional) A name to use as a name_scope for the operation.

Same as tensors.

ValueError If tensors does not contain any Tensor or IndexedSlices.
TypeError If control_inputs is not a list of Operation or Tensor objects.