1515from __future__ import annotations
1616
1717import logging
18+ import warnings
1819from copy import copy
1920from typing import Sequence
20-
2121import numpy as np
22+
2223from qiskit .circuit import Parameter , QuantumCircuit
23- from qiskit .primitives import BaseEstimator , Estimator , EstimatorResult
24+ from qiskit .primitives .base import BaseEstimatorV2
25+ from qiskit .primitives import BaseEstimator , BaseEstimatorV1 , Estimator , EstimatorResult
2426from qiskit .quantum_info import SparsePauliOp
2527from qiskit .quantum_info .operators .base_operator import BaseOperator
28+
2629from ..gradients import (
2730 BaseEstimatorGradient ,
2831 EstimatorGradientResult ,
2932 ParamShiftEstimatorGradient ,
3033)
31-
3234from ..circuit .library import QNNCircuit
3335from ..exceptions import QiskitMachineLearningError
3436
@@ -64,7 +66,7 @@ class EstimatorQNN(NeuralNetwork):
6466 num_qubits = 2
6567
6668 # Using the QNNCircuit:
67- # Create a parameterized 2 qubit circuit composed of the default ZZFeatureMap feature map
69+ # Create a parametrrized 2 qubit circuit composed of the default ZZFeatureMap feature map
6870 # and RealAmplitudes ansatz.
6971 qnn_qc = QNNCircuit(num_qubits)
7072
@@ -105,12 +107,14 @@ def __init__(
105107 self ,
106108 * ,
107109 circuit : QuantumCircuit ,
108- estimator : BaseEstimator | None = None ,
110+ estimator : BaseEstimator | BaseEstimatorV2 | None = None ,
109111 observables : Sequence [BaseOperator ] | BaseOperator | None = None ,
110112 input_params : Sequence [Parameter ] | None = None ,
111113 weight_params : Sequence [Parameter ] | None = None ,
112114 gradient : BaseEstimatorGradient | None = None ,
113115 input_gradients : bool = False ,
116+ num_virtual_qubits : int | None = None ,
117+ default_precision : float = 0.015625 ,
114118 ):
115119 r"""
116120 Args:
@@ -127,12 +131,12 @@ def __init__(
127131 input_params: The parameters that correspond to the input data of the network.
128132 If ``None``, the input data is not bound to any parameters.
129133 If a :class:`~qiskit_machine_learning.circuit.library.QNNCircuit` is provided the
130- `input_params` value here is ignored. Instead the value is taken from the
134+ `input_params` value here is ignored. Instead, the value is taken from the
131135 :class:`~qiskit_machine_learning.circuit.library.QNNCircuit` input_parameters.
132136 weight_params: The parameters that correspond to the trainable weights.
133137 If ``None``, the weights are not bound to any parameters.
134138 If a :class:`~qiskit_machine_learning.circuit.library.QNNCircuit` is provided the
135- `weight_params` value here is ignored. Instead the value is taken from the
139+ `weight_params` value here is ignored. Instead, the value is taken from the
136140 :class:`~qiskit_machine_learning.circuit.library.QNNCircuit` weight_parameters.
137141 gradient: The estimator gradient to be used for the backward pass.
138142 If None, a default instance of the estimator gradient,
@@ -141,6 +145,8 @@ def __init__(
141145 Note that this parameter is ``False`` by default, and must be explicitly set to
142146 ``True`` for a proper gradient computation when using
143147 :class:`~qiskit_machine_learning.connectors.TorchConnector`.
148+ num_virtual_qubits: Number of virtual qubits.
149+ default_precision: The default precision for the estimator if not specified during run.
144150
145151 Raises:
146152 QiskitMachineLearningError: Invalid parameter values.
@@ -149,19 +155,46 @@ def __init__(
149155 estimator = Estimator ()
150156 self .estimator = estimator
151157 self ._org_circuit = circuit
158+
159+ if num_virtual_qubits is None :
160+ self .num_virtual_qubits = circuit .num_qubits
161+ warnings .warn (
162+ f"No number of qubits was not specified ({ num_virtual_qubits }
163+ + f"`circuit` ({ self .num_virtual_qubits :d}
164+ + "unstable behaviour." ,
165+ UserWarning ,
166+ stacklevel = 2 ,
167+ )
168+ else :
169+ self .num_virtual_qubits = num_virtual_qubits
170+
152171 if observables is None :
153- observables = SparsePauliOp .from_list ([("Z" * circuit .num_qubits , 1 )])
172+ observables = SparsePauliOp .from_sparse_list (
173+ [("Z" * self .num_virtual_qubits , range (self .num_virtual_qubits ), 1 )],
174+ num_qubits = self .circuit .num_qubits ,
175+ )
176+
154177 if isinstance (observables , BaseOperator ):
155178 observables = (observables ,)
179+
156180 self ._observables = observables
181+
157182 if isinstance (circuit , QNNCircuit ):
158183 self ._input_params = list (circuit .input_parameters )
159184 self ._weight_params = list (circuit .weight_parameters )
160185 else :
161186 self ._input_params = list (input_params ) if input_params is not None else []
162187 self ._weight_params = list (weight_params ) if weight_params is not None else []
188+
163189 if gradient is None :
190+ if isinstance (self .estimator , BaseEstimatorV2 ):
191+ raise QiskitMachineLearningError (
192+ "Please provide a gradient with pass manager initialised."
193+ )
194+
164195 gradient = ParamShiftEstimatorGradient (self .estimator )
196+
197+ self ._default_precision = default_precision
165198 self .gradient = gradient
166199 self ._input_gradients = input_gradients
167200
@@ -198,33 +231,52 @@ def weight_params(self) -> Sequence[Parameter] | None:
198231 @property
199232 def input_gradients (self ) -> bool :
200233 """Returns whether gradients with respect to input data are computed by this neural network
201- in the ``backward`` method or not. By default such gradients are not computed."""
234+ in the ``backward`` method or not. By default, such gradients are not computed."""
202235 return self ._input_gradients
203236
204237 @input_gradients .setter
205238 def input_gradients (self , input_gradients : bool ) -> None :
206239 """Turn on/off computation of gradients with respect to input data."""
207240 self ._input_gradients = input_gradients
208241
242+ @property
243+ def default_precision (self ) -> float :
244+ """Return the default precision"""
245+ return self ._default_precision
246+
209247 def _forward_postprocess (self , num_samples : int , result : EstimatorResult ) -> np .ndarray :
210248 """Post-processing during forward pass of the network."""
211- return np .reshape (result . values , (- 1 , num_samples )).T
249+ return np .reshape (result , (- 1 , num_samples )).T
212250
213251 def _forward (
214252 self , input_data : np .ndarray | None , weights : np .ndarray | None
215253 ) -> np .ndarray | None :
216254 """Forward pass of the neural network."""
217255 parameter_values_ , num_samples = self ._preprocess_forward (input_data , weights )
218- job = self .estimator .run (
219- [self ._circuit ] * num_samples * self .output_shape [0 ],
220- [op for op in self ._observables for _ in range (num_samples )],
221- np .tile (parameter_values_ , (self .output_shape [0 ], 1 )),
222- )
223- try :
224- results = job .result ()
225- except Exception as exc :
226- raise QiskitMachineLearningError ("Estimator job failed." ) from exc
227256
257+ # Determine how to run the estimator based on its version
258+ if isinstance (self .estimator , BaseEstimatorV1 ):
259+ job = self .estimator .run (
260+ [self ._circuit ] * num_samples * self .output_shape [0 ],
261+ [op for op in self ._observables for _ in range (num_samples )],
262+ np .tile (parameter_values_ , (self .output_shape [0 ], 1 )),
263+ )
264+ results = job .result ().values
265+
266+ elif isinstance (self .estimator , BaseEstimatorV2 ):
267+ # Prepare circuit-observable-parameter tuples (PUBs)
268+ circuit_observable_params = []
269+ for observable in self ._observables :
270+ circuit_observable_params .append ((self ._circuit , observable , parameter_values_ ))
271+ # For BaseEstimatorV2, run the estimator using PUBs and specified precision
272+ job = self .estimator .run (circuit_observable_params , precision = self ._default_precision )
273+ results = [result .data .evs for result in job .result ()]
274+ else :
275+ raise QiskitMachineLearningError (
276+ "The accepted estimators are BaseEstimatorV1 and BaseEstimatorV2; got "
277+ + f"{ type (self .estimator )}
278+ + "Qiskit and removed in Qiskit IBM Runtime."
279+ )
228280 return self ._forward_postprocess (num_samples , results )
229281
230282 def _backward_postprocess (
@@ -269,8 +321,11 @@ def _backward(
269321 param_values = np .tile (parameter_values , (num_observables , 1 ))
270322
271323 job = None
324+
272325 if self ._input_gradients :
273- job = self .gradient .run (circuits , observables , param_values ) # type: ignore[arg-type]
326+ job = self .gradient .run (
327+ circuits , observables , param_values
328+ ) # type: ignore[arg-type]
274329 elif len (parameter_values [0 ]) > self ._num_inputs :
275330 params = [self ._circuit .parameters [self ._num_inputs :]] * num_circuits
276331 job = self .gradient .run (
@@ -281,7 +336,7 @@ def _backward(
281336 try :
282337 results = job .result ()
283338 except Exception as exc :
284- raise QiskitMachineLearningError ("Estimator job failed." ) from exc
339+ raise QiskitMachineLearningError (f "Estimator job failed. { exc } from exc
285340
286341 input_grad , weights_grad = self ._backward_postprocess (num_samples , results )
287342
0 commit comments