TF 2.0 is out! Get hands-on practice at TF World, Oct 28-31. Use code TF20 for 20% off select passes. Register now


Context manager for recording interceptable executions onto a tape.



Similar to tf.GradientTape, operations are recorded if they are executed within this context manager. In addition, the operation must be registered (wrapped) as ed.interceptable.


  • tape: OrderedDict where operations are recorded in sequence. Keys are the name keyword argument to the operation (typically, a random variable's name) and values are the corresponding output of the operation. If the operation has no name, it is not recorded.


from tensorflow_probability import edward2 as ed

def probabilistic_matrix_factorization():
  users = ed.Normal(0., 1., sample_shape=[5000, 128], name="users")
  items = ed.Normal(0., 1., sample_shape=[7500, 128], name="items")
  ratings = ed.Normal(loc=tf.matmul(users, items, transpose_b=True),
  return ratings

with ed.tape() as model_tape:
  ratings = probabilistic_matrix_factorization()

assert model_tape["users"].shape == (5000, 128)
assert model_tape["items"].shape == (7500, 128)
assert model_tape["ratings"] == ratings