Deprecate approximation option in Estimator (#1963)

* Deprecate approximation

* fix tests

qiskit_aer/primitives/estimator.py

@@ -32,6 +32,7 @@
 from qiskit.quantum_info import Pauli, PauliList
 from qiskit.quantum_info.operators.base_operator import BaseOperator
 from qiskit.result.models import ExperimentResult
+from qiskit.utils import deprecate_arg, deprecate_func
 
 from .. import AerError, AerSimulator
 
@@ -68,6 +69,12 @@ class Estimator(BaseEstimator):
           normal distribution approximation.
     """
 
+    @deprecate_arg(
+        "approximation",
+        since=0.13,
+        package_name="qiskit-aer",
+        additional_msg="approximation=True will be default in the future.",
+    )
     def __init__(
         self,
         *,
@@ -100,7 +107,15 @@ def __init__(
         self._transpile_options = Options()
         if transpile_options is not None:
             self._transpile_options.update_options(**transpile_options)
-        self.approximation = approximation
+        if not approximation:
+            warn(
+                "Option approximation=False is deprecated as of qiskit-aer 0.13. "
+                "It will be removed no earlier than 3 months after the release date. "
+                "Instead, use BackendEstmator from qiskit.primitives.",
+                DeprecationWarning,
+                stacklevel=3,
+            )
+        self._approximation = approximation
         self._skip_transpilation = skip_transpilation
         self._cache: dict[tuple[tuple[int], tuple[int], bool], tuple[dict, dict]] = {}
         self._transpiled_circuits: dict[int, QuantumCircuit] = {}
@@ -109,6 +124,34 @@ def __init__(
         self._observable_ids: dict[tuple, int] = {}
         self._abelian_grouping = abelian_grouping
 
+    @property
+    @deprecate_func(
+        since=0.13,
+        package_name="qiskit-aer",
+        is_property=True,
+    )
+    def approximation(self):
+        """The approximation property"""
+        return self._approximation
+
+    @approximation.setter
+    @deprecate_func(
+        since=0.13,
+        package_name="qiskit-aer",
+        is_property=True,
+    )
+    def approximation(self, approximation):
+        """Setter for approximation"""
+        if not approximation:
+            warn(
+                "Option approximation=False is deprecated as of qiskit-aer 0.13. "
+                "It will be removed no earlier than 3 months after the release date. "
+                "Instead, use BackendEstmator from qiskit.primitives.",
+                DeprecationWarning,
+                stacklevel=3,
+            )
+        self._approximation = approximation
+
     def _call(
         self,
         circuits: Sequence[int],
@@ -120,7 +163,7 @@ def _call(
         if seed is not None:
             run_options.setdefault("seed_simulator", seed)
 
-        if self.approximation:
+        if self._approximation:
             return self._compute_with_approximation(
                 circuits, observables, parameter_values, run_options, seed
             )
@@ -174,7 +217,7 @@ def _compute(self, circuits, observables, parameter_values, run_options):
             )
 
         # Key for cache
-        key = (tuple(circuits), tuple(observables), self.approximation)
+        key = (tuple(circuits), tuple(observables), self._approximation)
 
         # Create expectation value experiments.
         if key in self._cache:  # Use a cache
@@ -363,7 +406,7 @@ def _compute_with_approximation(
         self, circuits, observables, parameter_values, run_options, seed
     ):
         # Key for cache
-        key = (tuple(circuits), tuple(observables), self.approximation)
+        key = (tuple(circuits), tuple(observables), self._approximation)
         shots = run_options.pop("shots", None)
 
         # Create expectation value experiments.

test/terra/primitives/test_estimator.py

@@ -69,9 +69,10 @@ def test_estimator(self, abelian_grouping):
         with self.subTest("SparsePauliOp"):
             observable = SparsePauliOp.from_list(lst)
             ansatz = RealAmplitudes(num_qubits=2, reps=2)
-            est = Estimator(
-                backend_options={"method": "statevector"}, abelian_grouping=abelian_grouping
-            )
+            with self.assertWarns(DeprecationWarning):
+                est = Estimator(
+                    backend_options={"method": "statevector"}, abelian_grouping=abelian_grouping
+                )
             result = est.run(
                 ansatz, observable, parameter_values=[[0, 1, 1, 2, 3, 5]], seed=15
             ).result()
@@ -88,9 +89,10 @@ def test_estimator(self, abelian_grouping):
                 ]
             )
             ansatz = RealAmplitudes(num_qubits=2, reps=2)
-            est = Estimator(
-                backend_options={"method": "statevector"}, abelian_grouping=abelian_grouping
-            )
+            with self.assertWarns(DeprecationWarning):
+                est = Estimator(
+                    backend_options={"method": "statevector"}, abelian_grouping=abelian_grouping
+                )
             result = est.run(ansatz, observable, parameter_values=[[0] * 6], seed=15).result()
             self.assertIsInstance(result, EstimatorResult)
             np.testing.assert_allclose(result.values, [-0.4], rtol=0.02)
@@ -107,15 +109,17 @@ def test_init_observable_from_operator(self, abelian_grouping):
                 [0.1809312, 0.0, 0.0, -1.06365335],
             ]
         )
-        est = Estimator(abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(abelian_grouping=abelian_grouping)
         result = est.run([circuit], [matrix], seed=15, shots=8192).result()
         self.assertIsInstance(result, EstimatorResult)
         np.testing.assert_allclose(result.values, [self.expval], rtol=0.02)
 
     @data(True, False)
     def test_evaluate(self, abelian_grouping):
         """test for evaluate"""
-        est = Estimator(abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(abelian_grouping=abelian_grouping)
         result = est.run(
             self.ansatz, self.observable, parameter_values=[[0, 1, 1, 2, 3, 5]], seed=15, shots=8192
         ).result()
@@ -125,7 +129,8 @@ def test_evaluate(self, abelian_grouping):
     @data(True, False)
     def test_evaluate_multi_params(self, abelian_grouping):
         """test for evaluate with multiple parameters"""
-        est = Estimator(abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(abelian_grouping=abelian_grouping)
         result = est.run(
             [self.ansatz] * 2,
             [self.observable] * 2,
@@ -139,7 +144,8 @@ def test_evaluate_multi_params(self, abelian_grouping):
     def test_evaluate_no_params(self, abelian_grouping):
         """test for evaluate without parameters"""
         circuit = self.ansatz.assign_parameters([0, 1, 1, 2, 3, 5])
-        est = Estimator(abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(abelian_grouping=abelian_grouping)
         result = est.run(circuit, self.observable, seed=15, shots=8192).result()
         self.assertIsInstance(result, EstimatorResult)
         np.testing.assert_allclose(result.values, [self.expval], rtol=0.02)
@@ -151,7 +157,8 @@ def test_run_with_multiple_observables_and_none_parameters(self, abelian_groupin
         circuit.h(0)
         circuit.cx(0, 1)
         circuit.cx(1, 2)
-        est = Estimator(abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(abelian_grouping=abelian_grouping)
         result = est.run(
             [circuit] * 2, [SparsePauliOp("ZZZ"), SparsePauliOp("III")], seed=15
         ).result()
@@ -168,7 +175,8 @@ def test_1qubit(self, abelian_grouping):
         op0 = SparsePauliOp.from_list([("I", 1)])
         op1 = SparsePauliOp.from_list([("Z", 1)])
 
-        est = Estimator(abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(abelian_grouping=abelian_grouping)
         with self.subTest("test circuit 0, observable 0"):
             result = est.run(qc0, op0).result()
             self.assertIsInstance(result, EstimatorResult)
@@ -200,7 +208,8 @@ def test_2qubits(self, abelian_grouping):
         op1 = SparsePauliOp.from_list([("ZI", 1)])
         op2 = SparsePauliOp.from_list([("IZ", 1)])
 
-        est = Estimator(abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(abelian_grouping=abelian_grouping)
         with self.subTest("test circuit 0, observable 0"):
             result = est.run(qc0, op0).result()
             self.assertIsInstance(result, EstimatorResult)
@@ -237,7 +246,8 @@ def test_empty_parameter(self, abelian_grouping):
         n = 2
         qc = QuantumCircuit(n)
         op = SparsePauliOp.from_list([("I" * n, 1)])
-        estimator = Estimator(abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            estimator = Estimator(abelian_grouping=abelian_grouping)
         with self.subTest("one circuit"):
             result = estimator.run(qc, op, shots=1000).result()
             np.testing.assert_allclose(result.values, [1])
@@ -257,7 +267,8 @@ def test_numpy_params(self, abelian_grouping):
         params_array = np.random.rand(k, qc.num_parameters)
         params_list = params_array.tolist()
         params_list_array = list(params_array)
-        estimator = Estimator(abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            estimator = Estimator(abelian_grouping=abelian_grouping)
         target = estimator.run([qc] * k, [op] * k, params_list, seed=15).result()
 
         with self.subTest("ndarrary"):
@@ -275,7 +286,8 @@ def test_with_shots_option_with_approximation(self, abelian_grouping):
         """test with shots option."""
         # Note: abelian_gropuing is ignored when approximation is True as documented.
         # The purpose of this test is to make sure the results remain the same.
-        est = Estimator(approximation=True, abelian_grouping=abelian_grouping)
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(approximation=True, abelian_grouping=abelian_grouping)
         result = est.run(
             self.ansatz, self.observable, parameter_values=[[0, 1, 1, 2, 3, 5]], shots=1024, seed=15
         ).result()
@@ -285,7 +297,8 @@ def test_with_shots_option_with_approximation(self, abelian_grouping):
 
     def test_with_shots_option_without_approximation(self):
         """test with shots option."""
-        est = Estimator(approximation=False, abelian_grouping=False)
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(approximation=False, abelian_grouping=False)
         result = est.run(
             self.ansatz, self.observable, parameter_values=[[0, 1, 1, 2, 3, 5]], shots=1024, seed=15
         ).result()
@@ -295,15 +308,15 @@ def test_with_shots_option_without_approximation(self):
 
     def test_warn_shots_none_without_approximation(self):
         """Test waning for shots=None without approximation."""
-        est = Estimator(approximation=False)
-        with self.assertWarns(RuntimeWarning):
-            result = est.run(
-                self.ansatz,
-                self.observable,
-                parameter_values=[[0, 1, 1, 2, 3, 5]],
-                shots=None,
-                seed=15,
-            ).result()
+        with self.assertWarns(DeprecationWarning):
+            est = Estimator(approximation=False)
+        result = est.run(
+            self.ansatz,
+            self.observable,
+            parameter_values=[[0, 1, 1, 2, 3, 5]],
+            shots=None,
+            seed=15,
+        ).result()
         self.assertIsInstance(result, EstimatorResult)
         np.testing.assert_allclose(result.values, [-1.313831587508902])
         self.assertIsInstance(result.metadata[0]["variance"], float)
@@ -318,7 +331,8 @@ def test_result_order(self):
         qc2.ry(np.pi / 2 * param, 0)
         qc2.measure_all()
 
-        estimator = Estimator(approximation=True)
+        with self.assertWarns(DeprecationWarning):
+            estimator = Estimator(approximation=True)
         job = estimator.run([qc1, qc2, qc1, qc1, qc2], ["Z"] * 5, [[], [1], [], [], [1]])
         result = job.result()
         np.testing.assert_allclose(result.values, [1, 0, 1, 1, 0], atol=1e-10)