quantum_instance.py

# This code is part of Qiskit.
#
# (C) Copyright IBM 2018, 2021.
#
# This code is licensed under the Apache License, Version 2.0. You may
# obtain a copy of this license in the LICENSE.txt file in the root directory
# of this source tree or at http://www.apache.org/licenses/LICENSE-2.0.
#
# Any modifications or derivative works of this code must retain this
# copyright notice, and modified files need to carry a notice indicating
# that they have been altered from the originals.

""" Quantum Instance module """

from typing import Optional, List, Union, Dict, Callable, Tuple
import copy
import logging
import time
import numpy as np

from qiskit.providers import Backend, BaseBackend
from qiskit.transpiler import CouplingMap, PassManager
from qiskit.transpiler.layout import Layout
from qiskit.assembler.run_config import RunConfig
from qiskit.circuit import QuantumCircuit
from qiskit.result import Result
from qiskit.qobj import Qobj
from qiskit import compiler

try:
    from qiskit.providers.aer.noise import NoiseModel  # pylint: disable=unused-import
except ImportError as ex:
    pass

from .aqua_error import AquaError
from .utils.backend_utils import (is_ibmq_provider,
                                  is_statevector_backend,
                                  is_simulator_backend,
                                  is_local_backend,
                                  is_aer_qasm,
                                  is_basicaer_provider,
                                  support_backend_options)
from .utils.circuit_utils import summarize_circuits

logger = logging.getLogger(__name__)


class QuantumInstance:
    """Quantum Backend including execution setting."""

    _BACKEND_CONFIG = ['basis_gates', 'coupling_map']
    _COMPILE_CONFIG = ['initial_layout', 'seed_transpiler', 'optimization_level']
    _RUN_CONFIG = ['shots', 'max_credits', 'memory', 'seed_simulator']
    _QJOB_CONFIG = ['timeout', 'wait']
    _NOISE_CONFIG = ['noise_model']

    # https://github.com/Qiskit/qiskit-aer/blob/master/qiskit/providers/aer/backends/qasm_simulator.py
    _BACKEND_OPTIONS_QASM_ONLY = ["statevector_sample_measure_opt", "max_parallel_shots"]
    _BACKEND_OPTIONS = ["initial_statevector", "chop_threshold", "max_parallel_threads",
                        "max_parallel_experiments", "statevector_parallel_threshold",
                        "statevector_hpc_gate_opt"] + _BACKEND_OPTIONS_QASM_ONLY

    def __init__(self,
                 backend: Union[Backend, BaseBackend],
                 # run config
                 shots: int = 1024,
                 seed_simulator: Optional[int] = None,
                 max_credits: int = 10,
                 # backend properties
                 basis_gates: Optional[List[str]] = None,
                 coupling_map: Optional[Union[CouplingMap, List[List]]] = None,
                 # transpile
                 initial_layout: Optional[Union[Layout, Dict, List]] = None,
                 pass_manager: Optional[PassManager] = None,
                 seed_transpiler: Optional[int] = None,
                 optimization_level: Optional[int] = None,
                 # simulation
                 backend_options: Optional[Dict] = None,
                 noise_model: Optional['NoiseModel'] = None,
                 # job
                 timeout: Optional[float] = None,
                 wait: float = 5.,
                 # others
                 skip_qobj_validation: bool = True,
                 measurement_error_mitigation_cls: Optional[Callable] = None,
                 cals_matrix_refresh_period: int = 30,
                 measurement_error_mitigation_shots: Optional[int] = None,
                 job_callback: Optional[Callable] = None) -> None:
        """
        Quantum Instance holds a Qiskit Terra backend as well as configuration for circuit
        transpilation and execution. When provided to an Aqua algorithm the algorithm will
        execute the circuits it needs to run using the instance.

        Args:
            backend: Instance of selected backend
            shots: Number of repetitions of each circuit, for sampling
            seed_simulator: Random seed for simulators
            max_credits: Maximum credits to use
            basis_gates: List of basis gate names supported by the
                                               target. Defaults to basis gates of the backend.
            coupling_map: Coupling map (perhaps custom) to
                                                      target in mapping
            initial_layout: Initial layout of qubits in mapping
            pass_manager: Pass manager to handle how to compile the circuits
            seed_transpiler: The random seed for circuit mapper
            optimization_level: How much optimization to perform on the circuits.
                Higher levels generate more optimized circuits, at the expense of longer
                transpilation time.
            backend_options: All running options for backend, please refer
                to the provider of the backend for information as to what options it supports.
            noise_model: noise model for simulator
            timeout: Seconds to wait for job. If None, wait indefinitely.
            wait: Seconds between queries for job result
            skip_qobj_validation: Bypass Qobj validation to decrease circuit
                processing time during submission to backend.
            measurement_error_mitigation_cls: The approach to mitigate
                measurement errors. Qiskit Ignis provides fitter classes for this functionality
                and CompleteMeasFitter from qiskit.ignis.mitigation.measurement module can be used
                here. (TensoredMeasFitter is not supported).
            cals_matrix_refresh_period: How often to refresh the calibration
                matrix in measurement mitigation. in minutes
            measurement_error_mitigation_shots: The number of shots number for
                building calibration matrix. If None, the main `shots` parameter value is used.
            job_callback: Optional user supplied callback which can be used
                to monitor job progress as jobs are submitted for processing by an Aqua algorithm.
                The callback is provided the following arguments: `job_id, job_status,
                queue_position, job`

        Raises:
            AquaError: the shots exceeds the maximum number of shots
            AquaError: set noise model but the backend does not support that
            AquaError: set backend_options but the backend does not support that
        """
        from .deprecation import warn_class
        warn_class('aqua.QuantumInstance',
                   'qiskit.utils.QuantumInstance',
                   'qiskit-terra')
        self._backend = backend
        self._pass_manager = pass_manager

        # setup run config
        if shots is not None:
            if self.is_statevector and shots != 1:
                logger.info("statevector backend only works with shot=1, changing "
                            "shots from %s to 1.", shots)
                shots = 1
            max_shots = self._backend.configuration().max_shots
            if max_shots is not None and shots > max_shots:
                raise AquaError('The maximum shots supported by the selected backend is {} '
                                'but you specified {}'.format(max_shots, shots))

        run_config = RunConfig(shots=shots, max_credits=max_credits)
        if seed_simulator is not None:
            run_config.seed_simulator = seed_simulator

        self._run_config = run_config

        # setup backend config
        basis_gates = basis_gates or backend.configuration().basis_gates
        coupling_map = coupling_map or getattr(backend.configuration(), 'coupling_map', None)
        self._backend_config = {
            'basis_gates': basis_gates,
            'coupling_map': coupling_map
        }

        # setup compile config
        self._compile_config = {
            'initial_layout': initial_layout,
            'seed_transpiler': seed_transpiler,
            'optimization_level': optimization_level
        }

        # setup job config
        self._qjob_config = {'timeout': timeout} if self.is_local \
            else {'timeout': timeout, 'wait': wait}

        # setup noise config
        self._noise_config = {}
        if noise_model is not None:
            if is_simulator_backend(self._backend) and not is_basicaer_provider(self._backend):
                self._noise_config = {'noise_model': noise_model}
            else:
                raise AquaError("The noise model is not supported on the selected backend {} ({}) "
                                "only certain backends, such as Aer qasm simulator "
                                "support noise.".format(self.backend_name,
                                                        self._backend.provider()))

        # setup backend options for run
        self._backend_options = {}
        if backend_options is not None:
            if support_backend_options(self._backend):
                self._backend_options = {'backend_options': backend_options}
            else:
                raise AquaError("backend_options can not used with the backends in IBMQ provider.")

        # setup measurement error mitigation
        self._meas_error_mitigation_cls = None
        if self.is_statevector:
            if measurement_error_mitigation_cls is not None:
                raise AquaError("Measurement error mitigation does not work "
                                "with the statevector simulation.")
        else:
            self._meas_error_mitigation_cls = measurement_error_mitigation_cls
        self._meas_error_mitigation_fitters: Dict[str, Tuple[np.ndarray, float]] = {}
        # TODO: support different fitting method in error mitigation?
        self._meas_error_mitigation_method = 'least_squares'
        self._cals_matrix_refresh_period = cals_matrix_refresh_period
        self._meas_error_mitigation_shots = measurement_error_mitigation_shots

        if self._meas_error_mitigation_cls is not None:
            logger.info("The measurement error mitigation is enabled. "
                        "It will automatically submit an additional job to help "
                        "calibrate the result of other jobs. "
                        "The current approach will submit a job with 2^N circuits "
                        "to build the calibration matrix, "
                        "where N is the number of measured qubits. "
                        "Furthermore, Aqua will re-use the calibration matrix for %s minutes "
                        "and re-build it after that.", self._cals_matrix_refresh_period)

        # setup others
        if is_ibmq_provider(self._backend):
            if skip_qobj_validation:
                logger.info("skip_qobj_validation was set True but this setting is not "
                            "supported by IBMQ provider and has been ignored.")
                skip_qobj_validation = False
        self._skip_qobj_validation = skip_qobj_validation
        self._circuit_summary = False
        self._job_callback = job_callback
        self._time_taken = 0.
        logger.info(self)

    def __str__(self) -> str:
        """Overload string.

        Returns:
            str: the info of the object.
        """
        # pylint: disable=import-outside-toplevel
        from qiskit import __version__ as terra_version

        info = "\nQiskit Terra version: {}\n".format(terra_version)
        info += "Backend: '{} ({})', with following setting:\n{}\n{}\n{}\n{}\n{}\n{}".format(
            self.backend_name, self._backend.provider(), self._backend_config, self._compile_config,
            self._run_config, self._qjob_config, self._backend_options, self._noise_config)
        info += "\nMeasurement mitigation: {}".format(self._meas_error_mitigation_cls)

        return info

    def transpile(self,
                  circuits: Union[QuantumCircuit, List[QuantumCircuit]]) -> List[QuantumCircuit]:
        """
        A wrapper to transpile circuits to allow algorithm access the transpiled circuits.
        Args:
            circuits: circuits to transpile
        Returns:
            The transpiled circuits, it is always a list even though the length is one.
        """
        if self._pass_manager is not None:
            transpiled_circuits = self._pass_manager.run(circuits)
        else:
            transpiled_circuits = compiler.transpile(circuits,
                                                     self._backend,
                                                     **self._backend_config,
                                                     **self._compile_config)
        if not isinstance(transpiled_circuits, list):
            transpiled_circuits = [transpiled_circuits]

        if logger.isEnabledFor(logging.DEBUG) and self._circuit_summary:
            logger.debug("==== Before transpiler ====")
            logger.debug(summarize_circuits(circuits))
            if transpiled_circuits is not None:
                logger.debug("====  After transpiler ====")
                logger.debug(summarize_circuits(transpiled_circuits))

        return transpiled_circuits

    def assemble(self,
                 circuits: Union[QuantumCircuit, List[QuantumCircuit]]) -> Qobj:
        """ assemble circuits """
        return compiler.assemble(circuits, **self._run_config.to_dict())

    def execute(self,
                circuits: Union[QuantumCircuit, List[QuantumCircuit]],
                had_transpiled: bool = False) -> Result:
        """
        A wrapper to interface with quantum backend.

        Args:
            circuits: circuits to execute
            had_transpiled: whether or not circuits had been transpiled

        Returns:
            Result object

        TODO: Maybe we can combine the circuits for the main ones and calibration circuits before
              assembling to the qobj.
        """
        # pylint: disable=import-outside-toplevel
        from .utils.run_circuits import run_qobj

        from .utils.measurement_error_mitigation import (get_measured_qubits_from_qobj,
                                                         build_measurement_error_mitigation_qobj)
        # maybe compile
        if not had_transpiled:
            circuits = self.transpile(circuits)

        # assemble
        qobj = self.assemble(circuits)

        if self._meas_error_mitigation_cls is not None:
            qubit_index, qubit_mappings = get_measured_qubits_from_qobj(qobj)
            qubit_index_str = '_'.join([str(x) for x in qubit_index]) + \
                "_{}".format(self._meas_error_mitigation_shots or self._run_config.shots)
            meas_error_mitigation_fitter, timestamp = \
                self._meas_error_mitigation_fitters.get(qubit_index_str, (None, 0.))

            if meas_error_mitigation_fitter is None:
                # check the asked qubit_index are the subset of build matrices
                for key, _ in self._meas_error_mitigation_fitters.items():
                    stored_qubit_index = [int(x) for x in key.split("_")[:-1]]
                    stored_shots = int(key.split("_")[-1])
                    if len(qubit_index) < len(stored_qubit_index):
                        tmp = list(set(qubit_index + stored_qubit_index))
                        if sorted(tmp) == sorted(stored_qubit_index) and \
                                self._run_config.shots == stored_shots:
                            # the qubit used in current job is the subset and shots are the same
                            meas_error_mitigation_fitter, timestamp = \
                                self._meas_error_mitigation_fitters.get(key, (None, 0.))
                            meas_error_mitigation_fitter = \
                                meas_error_mitigation_fitter.subset_fitter(  # type: ignore
                                    qubit_sublist=qubit_index)
                            logger.info("The qubits used in the current job is the subset of "
                                        "previous jobs, "
                                        "reusing the calibration matrix if it is not out-of-date.")

            build_cals_matrix = self.maybe_refresh_cals_matrix(timestamp) or \
                meas_error_mitigation_fitter is None

            if build_cals_matrix:
                logger.info("Updating qobj with the circuits for measurement error mitigation.")
                use_different_shots = not (
                    self._meas_error_mitigation_shots is None
                    or self._meas_error_mitigation_shots == self._run_config.shots)
                temp_run_config = copy.deepcopy(self._run_config)
                if use_different_shots:
                    temp_run_config.shots = self._meas_error_mitigation_shots

                cals_qobj, state_labels, circuit_labels = \
                    build_measurement_error_mitigation_qobj(qubit_index,
                                                            self._meas_error_mitigation_cls,
                                                            self._backend,
                                                            self._backend_config,
                                                            self._compile_config,
                                                            temp_run_config)
                if use_different_shots or is_aer_qasm(self._backend):
                    cals_result = run_qobj(cals_qobj, self._backend, self._qjob_config,
                                           self._backend_options,
                                           self._noise_config,
                                           self._skip_qobj_validation, self._job_callback)
                    self._time_taken += cals_result.time_taken
                    result = run_qobj(qobj, self._backend, self._qjob_config,
                                      self._backend_options, self._noise_config,
                                      self._skip_qobj_validation, self._job_callback)
                    self._time_taken += result.time_taken
                else:
                    # insert the calibration circuit into main qobj if the shots are the same
                    qobj.experiments[0:0] = cals_qobj.experiments
                    result = run_qobj(qobj, self._backend, self._qjob_config,
                                      self._backend_options, self._noise_config,
                                      self._skip_qobj_validation, self._job_callback)
                    self._time_taken += result.time_taken
                    cals_result = result

                logger.info("Building calibration matrix for measurement error mitigation.")
                meas_error_mitigation_fitter = \
                    self._meas_error_mitigation_cls(cals_result,
                                                    state_labels,
                                                    qubit_list=qubit_index,
                                                    circlabel=circuit_labels)
                self._meas_error_mitigation_fitters[qubit_index_str] = \
                    (meas_error_mitigation_fitter, time.time())
            else:
                result = run_qobj(qobj, self._backend, self._qjob_config,
                                  self._backend_options, self._noise_config,
                                  self._skip_qobj_validation, self._job_callback)
                self._time_taken += result.time_taken

            if meas_error_mitigation_fitter is not None:
                logger.info("Performing measurement error mitigation.")
                skip_num_circuits = len(result.results) - len(circuits)
                #  remove the calibration counts from result object to assure the length of
                #  ExperimentalResult is equal length to input circuits
                result.results = result.results[skip_num_circuits:]
                tmp_result = copy.deepcopy(result)
                for qubit_index_str, c_idx in qubit_mappings.items():
                    curr_qubit_index = [int(x) for x in qubit_index_str.split("_")]
                    tmp_result.results = [result.results[i] for i in c_idx]
                    if curr_qubit_index == qubit_index:
                        tmp_fitter = meas_error_mitigation_fitter
                    else:
                        tmp_fitter = \
                            meas_error_mitigation_fitter.subset_fitter(  # type: ignore
                                curr_qubit_index)
                    tmp_result = tmp_fitter.filter.apply(  # type: ignore
                        tmp_result, self._meas_error_mitigation_method
                    )
                    for i, n in enumerate(c_idx):
                        result.results[n] = tmp_result.results[i]

        else:
            result = run_qobj(qobj, self._backend, self._qjob_config,
                              self._backend_options, self._noise_config,
                              self._skip_qobj_validation, self._job_callback)
            self._time_taken += result.time_taken

        if self._circuit_summary:
            self._circuit_summary = False

        return result

    def set_config(self, **kwargs):
        """Set configurations for the quantum instance."""
        for k, v in kwargs.items():
            if k in QuantumInstance._RUN_CONFIG:
                setattr(self._run_config, k, v)
            elif k in QuantumInstance._QJOB_CONFIG:
                self._qjob_config[k] = v
            elif k in QuantumInstance._COMPILE_CONFIG:
                self._compile_config[k] = v
            elif k in QuantumInstance._BACKEND_CONFIG:
                self._backend_config[k] = v
            elif k in QuantumInstance._BACKEND_OPTIONS:
                if not support_backend_options(self._backend):
                    raise AquaError("backend_options can not be used with this backend "
                                    "{} ({}).".format(self.backend_name, self._backend.provider()))

                if k in QuantumInstance._BACKEND_OPTIONS_QASM_ONLY and self.is_statevector:
                    raise AquaError("'{}' is only applicable for qasm simulator but "
                                    "statevector simulator is used as the backend.")

                if 'backend_options' not in self._backend_options:
                    self._backend_options['backend_options'] = {}
                self._backend_options['backend_options'][k] = v
            elif k in QuantumInstance._NOISE_CONFIG:
                if not is_simulator_backend(self._backend) or is_basicaer_provider(self._backend):
                    raise AquaError(
                        "The noise model is not supported on the selected backend {} ({}) "
                        "only certain backends, such as Aer qasm support "
                        "noise.".format(self.backend_name, self._backend.provider()))

                self._noise_config[k] = v

            else:
                raise ValueError("unknown setting for the key ({}).".format(k))

    @property
    def time_taken(self) -> float:
        """Accumulated time taken for execution."""
        return self._time_taken

    def reset_execution_results(self) -> None:
        """ Reset execution results """
        self._time_taken = 0.

    @property
    def qjob_config(self):
        """Getter of qjob_config."""
        return self._qjob_config

    @property
    def backend_config(self):
        """Getter of backend_config."""
        return self._backend_config

    @property
    def compile_config(self):
        """Getter of compile_config."""
        return self._compile_config

    @property
    def run_config(self):
        """Getter of run_config."""
        return self._run_config

    @property
    def noise_config(self):
        """Getter of noise_config."""
        return self._noise_config

    @property
    def backend_options(self):
        """Getter of backend_options."""
        return self._backend_options

    @property
    def circuit_summary(self):
        """Getter of circuit summary."""
        return self._circuit_summary

    @circuit_summary.setter
    def circuit_summary(self, new_value):
        """ sets circuit summary """
        self._circuit_summary = new_value

    @property
    def measurement_error_mitigation_cls(self):  # pylint: disable=invalid-name
        """ returns measurement error mitigation cls """
        return self._meas_error_mitigation_cls

    @measurement_error_mitigation_cls.setter
    def measurement_error_mitigation_cls(self, new_value):  # pylint: disable=invalid-name
        """ sets measurement error mitigation cls """
        self._meas_error_mitigation_cls = new_value

    @property
    def cals_matrix_refresh_period(self):
        """ returns matrix refresh period """
        return self._cals_matrix_refresh_period

    @cals_matrix_refresh_period.setter
    def cals_matrix_refresh_period(self, new_value):
        """ sets matrix refresh period """
        self._cals_matrix_refresh_period = new_value

    @property
    def measurement_error_mitigation_shots(self):  # pylint: disable=invalid-name
        """ returns measurement error mitigation shots """
        return self._meas_error_mitigation_shots

    @measurement_error_mitigation_shots.setter
    def measurement_error_mitigation_shots(self, new_value):  # pylint: disable=invalid-name
        """ sets measurement error mitigation shots """
        self._meas_error_mitigation_shots = new_value

    @property
    def backend(self):
        """Return BaseBackend backend object."""
        return self._backend

    @property
    def backend_name(self):
        """Return backend name."""
        return self._backend.name()

    @property
    def is_statevector(self):
        """Return True if backend is a statevector-type simulator."""
        return is_statevector_backend(self._backend)

    @property
    def is_simulator(self):
        """Return True if backend is a simulator."""
        return is_simulator_backend(self._backend)

    @property
    def is_local(self):
        """Return True if backend is a local backend."""
        return is_local_backend(self._backend)

    @property
    def skip_qobj_validation(self):
        """ checks if skip qobj validation """
        return self._skip_qobj_validation

    @skip_qobj_validation.setter
    def skip_qobj_validation(self, new_value):
        """ sets skip qobj validation flag """
        self._skip_qobj_validation = new_value

    def maybe_refresh_cals_matrix(self,
                                  timestamp: Optional[float] = None) -> bool:
        """
        Calculate the time difference from the query of last time.

        Args:
            timestamp: timestamp

        Returns:
            Whether or not refresh the cals_matrix
        """
        timestamp = timestamp or 0.
        ret = False
        curr_timestamp = time.time()
        difference = int(curr_timestamp - timestamp) / 60.0
        if difference > self._cals_matrix_refresh_period:
            ret = True

        return ret

    def cals_matrix(self,
                    qubit_index: Optional[List[int]] = None) -> \
            Optional[Union[Tuple[np.ndarray, float], Dict[str, Tuple[np.ndarray, float]]]]:
        """
        Get the stored calibration matrices and its timestamp.

        Args:
            qubit_index: the qubit index of corresponding calibration matrix.
                         If None, return all stored calibration matrices.

        Returns:
            The calibration matrix and the creation timestamp if qubit_index
            is not None otherwise, return all matrices and their timestamp
            in a dictionary.
        """
        shots = self._meas_error_mitigation_shots or self._run_config.shots
        if qubit_index:
            qubit_index_str = '_'.join([str(x) for x in qubit_index]) + "_{}".format(shots)
            fitter, timestamp = self._meas_error_mitigation_fitters.get(qubit_index_str, None)
            if fitter is not None:
                return fitter.cal_matrix, timestamp  # type: ignore
        else:
            return {k: (v.cal_matrix, t) for k, (v, t)  # type: ignore
                    in self._meas_error_mitigation_fitters.items()}
        return None