use TF 2.13 in CI (#4472)

* use TF 2.13 in CI

* fix requires_grad; stop using obs hash

* fix import; consider ndim=None to be an error

* changelog

* pull wires out of evil test

* make sure 2.12 still works

* Revert "make sure 2.12 still works"

This reverts commit 51d5e80.

* Update pennylane/math/single_dispatch.py

Co-authored-by: Christina Lee <christina@xanadu.ai>

* fix jax seed in test

---------

Co-authored-by: Christina Lee <christina@xanadu.ai>

.github/workflows/interface-unit-tests.yml

@@ -15,7 +15,7 @@ on:
  description: The version of TensorFlow to install for any job that requires TensorFlow
  required: false
  type: string
- default: 2.12.0
+ default: 2.13.0
  pytorch_version:
  description: The version of PyTorch to install for any job that requires PyTorch
  required: false

doc/releases/changelog-dev.md

@@ -227,6 +227,9 @@ array([False, False])
  [(#4465)](https://github.com/PennyLaneAI/pennylane/pull/4465/)
  [(#4478)](https://github.com/PennyLaneAI/pennylane/pull/4478/)
 
+* CI now runs tests with Tensorflow 2.13.0
+ [(#4472)](https://github.com/PennyLaneAI/pennylane/pull/4472)
+
 <h3>Breaking changes 💔</h3>
 
 * Gradient transforms no longer implicitly cast `float32` parameters to `float64`. Finite diff

pennylane/gradients/hamiltonian_grad.py

@@ -25,11 +25,11 @@ def hamiltonian_grad(tape, idx):
  idx (int): index of parameter that we differentiate with respect to
  """
 
- op, _, p_idx = tape.get_operation(idx)
+ op, m_pos, p_idx = tape.get_operation(idx)
  new_tape = tape.copy(copy_operations=True)
 
  # get position in queue
- queue_position = tape.observables.index(op)
+ queue_position = m_pos - len(tape.operations)
  new_tape._measurements[queue_position] = qml.expval(op.ops[p_idx])
 
  new_tape._par_info = {}

pennylane/math/single_dispatch.py

@@ -295,7 +295,10 @@ def _round_tf(tensor, decimals=0):
 
 def _ndim_tf(tensor):
  try:
- return _i("tf").experimental.numpy.ndim(tensor)
+ ndim = _i("tf").experimental.numpy.ndim(tensor)
+ if ndim is None:
+ return len(tensor.shape)
+ return ndim
  except AttributeError:
  return len(tensor.shape)
 

pennylane/math/utils.py

@@ -455,7 +455,10 @@ def requires_grad(tensor, interface=None):
  import tensorflow as tf
 
  try:
- from tensorflow.python.eager.tape import should_record_backprop
+ try:
+ from tensorflow.python.eager.record import should_record_backprop
+ except ImportError: # pragma: no cover
+ from tensorflow.python.eager.tape import should_record_backprop
  except ImportError: # pragma: no cover
  from tensorflow.python.eager.tape import should_record as should_record_backprop
 
@@ -508,7 +511,10 @@ def in_backprop(tensor, interface=None):
  import tensorflow as tf
 
  try:
- from tensorflow.python.eager.tape import should_record_backprop
+ try:
+ from tensorflow.python.eager.record import should_record_backprop
+ except ImportError: # pragma: no cover
+ from tensorflow.python.eager.tape import should_record_backprop
  except ImportError: # pragma: no cover
  from tensorflow.python.eager.tape import should_record as should_record_backprop
 

pennylane/ops/qubit/hamiltonian.py

@@ -44,17 +44,20 @@ def _compute_grouping_indices(observables, grouping_type="qwc", method="rlf"):
 
  indices = []
  available_indices = list(range(len(observables)))
- for partition in observable_groups:
+ for partition in observable_groups: # pylint:disable=too-many-nested-blocks
  indices_this_group = []
  for pauli_word in partition:
  # find index of this pauli word in remaining original observables,
  for observable in observables:
  if qml.pauli.are_identical_pauli_words(pauli_word, observable):
- ind = observables.index(observable)
- indices_this_group.append(available_indices[ind])
- # delete this observable and its index, so it cannot be found again
- observables.pop(ind)
- available_indices.pop(ind)
+ for ind, obs in enumerate(observables):
+ if obs is not observable:
+ continue
+ indices_this_group.append(available_indices[ind])
+ # delete this observable and its index, so it cannot be found again
+ observables.pop(ind)
+ available_indices.pop(ind)
+ break
  break
  indices.append(tuple(indices_this_group))
 

pennylane/pauli/grouping/group_observables.py

@@ -242,16 +242,19 @@ def group_observables(observables, coefficients=None, grouping_type="qwc", metho
  # we cannot delete elements from the coefficients tensor, so we
  # use a proxy list memorising the indices for this logic
  coeff_indices = list(range(qml.math.shape(coefficients)[0]))
- for i, partition in enumerate(partitioned_paulis):
+ for i, partition in enumerate(partitioned_paulis): # pylint:disable=too-many-nested-blocks
  indices = []
  for pauli_word in partition:
  # find index of this pauli word in remaining original observables,
  for observable in observables:
  if are_identical_pauli_words(pauli_word, observable):
- ind = observables.index(observable)
- indices.append(coeff_indices[ind])
- observables.pop(ind)
- coeff_indices.pop(ind)
+ for ind, obs in enumerate(observables):
+ if obs is not observable:
+ continue
+ indices.append(coeff_indices[ind])
+ observables.pop(ind)
+ coeff_indices.pop(ind)
+ break
  break
 
  # add a tensor of coefficients to the grouped coefficients

tests/ops/qubit/test_parametric_ops.py

@@ -3766,12 +3766,13 @@ def test_simplify_rotations_grad_tf_function(self):
  import tensorflow as tf
 
  op = qml.U2
+ wires = list(range(op.num_wires))
 
  dev = qml.device("default.qubit", wires=2)
 
  @tf.function
  @qml.qnode(dev)
- def circuit(simplify, wires, *params, **hyperparams):
+ def circuit(simplify, *params, **hyperparams):
  if simplify:
  qml.simplify(op(*params, wires=wires, **hyperparams))
  else:
@@ -3780,19 +3781,19 @@ def circuit(simplify, wires, *params, **hyperparams):
  return qml.expval(qml.PauliZ(0))
 
  unsimplified_op = self.get_unsimplified_op(op)
- params, wires = self._get_params_wires(unsimplified_op)
+ params, _ = self._get_params_wires(unsimplified_op)
  hyperparams = {"dim": 2} if unsimplified_op.name == "PCPhase" else {}
 
  for i in range(params[0].shape[0]):
  parameters = [tf.Variable(p[i]) for p in params]
 
  with tf.GradientTape() as unsimplified_tape:
- unsimplified_res = circuit(False, wires, *parameters, **hyperparams)
+ unsimplified_res = circuit(False, *parameters, **hyperparams)
 
  unsimplified_grad = unsimplified_tape.gradient(unsimplified_res, parameters)
 
  with tf.GradientTape() as simplified_tape:
- simplified_res = circuit(True, wires, *parameters, **hyperparams)
+ simplified_res = circuit(True, *parameters, **hyperparams)
 
  simplified_grad = simplified_tape.gradient(simplified_res, parameters)
 

tests/test_operation.py

@@ -1989,8 +1989,8 @@ def test_map_wires(self):
  qml.Identity(1),
  cv.NumberOperator(wires=[1]),
  cv.TensorN(wires=[1]),
- cv.X(wires=[1]),
- cv.P(wires=[1]),
+ cv.QuadX(wires=[1]),
+ cv.QuadP(wires=[1]),
  # cv.QuadOperator(1.234, wires=0),
  # cv.FockStateProjector([1,2,3], wires=[0, 1, 2]),
  cv.PolyXP(np.array([1.0, 2.0, 3.0]), wires=[0]),