redo expressivity proxy

src/all_qas.py

@@ -1,7 +1,13 @@
 #!/usr/bin/env python
 from argparse import ArgumentParser
+from dataclasses import dataclass
 from enum import Enum
 
+from ga_qas import GeneticAlgorithmSettings
+from qd_qas import QualityDiversitySettings
+from settings import QuantumArchitectureSearchSettings
+from tf_qas import TrainingFreeSettings
+
 
 class SearchStrategy(Enum):
     TFQAS = 1
@@ -9,8 +15,7 @@ class SearchStrategy(Enum):
     QDQAS = 3
 
 
-def main(search_strategy: SearchStrategy):
-
+def main(search_settings: QualityDiversitySettings | TrainingFreeSettings | GeneticAlgorithmSettings):
     pass
 
 

src/ga_qas/__init__.py

@@ -0,0 +1,8 @@
+from dataclasses import dataclass
+
+from settings import QuantumArchitectureSearchSettings
+
+
+@dataclass
+class QualityDiversitySettings(QuantumArchitectureSearchSettings):
+    initial_population_size: int

src/qd_qas/__init__.py

@@ -0,0 +1,28 @@
+from dataclasses import dataclass
+from typing import Callable
+
+from settings import QuantumArchitectureSearchSettings
+
+
+@dataclass
+class QualityDiversitySettings(QuantumArchitectureSearchSettings):
+    """
+    Hyperparameters for Quality Diversity based Quantum Architecture Search
+    """
+
+    initial_population_size: int
+    """How many (random) circuits should be generated at the beginning"""
+
+    offspring_size: int
+    """How many offspring should be generated from the previous generation"""
+
+    generation_size: int
+    """How many of the previous generation offspring become a parent for the next generation"""
+
+    generation_count: int
+    """How many generations to optimize for"""
+
+    mutation_rate: float
+    """for each gate in a circuit from the offspring perform a mutation if random [0, 1) < mutation_rate"""
+
+    cost_function: Callable[[ParametrizedQuantumCircuit], float]

src/quantum_circuit/__init__.py

@@ -0,0 +1,129 @@
+import enum
+from dataclasses import dataclass
+
+from qiskit import QuantumCircuit
+from qiskit.circuit.parametervector import ParameterVector
+
+from quantum_circuit.proxy_config import ProxyConfig
+from quantum_circuit.qiskit_helpers import build_qiskit_circ
+
+from .proxies import (calculate_entanglement, calculate_expressivity,
+                      calculate_fidelity)
+
+
+class QuantumType(enum.Enum):
+    Identity = enum.auto()
+    Hadamard = enum.auto()
+    X = enum.auto()
+    RX = enum.auto()
+    RXX = enum.auto()
+    Y = enum.auto()
+    RY = enum.auto()
+    RYY = enum.auto()
+    Z = enum.auto()
+    RZ = enum.auto()
+    RZZ = enum.auto()
+    CX = enum.auto()
+    CRX = enum.auto()
+    CZ = enum.auto()
+
+    def is_single_qubit(self):
+        return self in {
+            QuantumType.Identity,
+            QuantumType.Hadamard,
+            QuantumType.X,
+            QuantumType.RX,
+            QuantumType.Y,
+            QuantumType.RY,
+            QuantumType.Z,
+            QuantumType.RZ,
+        }
+
+
+@dataclass(frozen=True)
+class QuantumGate:
+    type: QuantumType
+    qubits: tuple[int] | tuple[int, int]
+
+
+class ParametrizedQuantumCircuit:
+    """
+    A Quantum circuit representation to use in the Quantum Architecture Searches.
+    It has methods for various proxies and stores the results as well, to have everything close.
+    Allows easy access to the gates inside so it can be used by functions that use it.
+    """
+
+    qubits: int
+    """How many qubits does the circuit contain"""
+
+    gates: list[list[QuantumGate]]
+    """Gates in the quantum circuit, organised as a list of layers of gates."""
+
+    _expressivity: float | None = None
+    """
+    Expressivity of the circuit, following the definition from {THE PAPER}
+    Use `expressivity` to get the value
+    """  # TODO: link paper
+
+    _entanglement: float | None = None
+    """
+    Entanglement of the circuit, following the definition from {THE PAPER}
+    Use `entanglement` to get the value
+    """  # TODO: link paper
+
+    _fidelity: float | None = None
+    """
+    Approximated fidelity of the circuit
+    Use `fidelity` to get the value
+    """
+
+    _qiskit_circ: tuple[QuantumCircuit, ParameterVector] | None = None
+
+    def __init__(self, qubits: int):
+        """
+        Initialise an empty circuit for the passed amount of qubits
+        """
+        self.qubits = qubits
+        self.gates = []
+
+    def append_layer(self, layer: list[QuantumGate]):
+        """
+        Add a layer to the end of the existing circuit in-place
+        """
+        self.gates.append(layer)
+
+    @property
+    def circ(self) -> tuple[QuantumCircuit, ParameterVector]:
+        """
+        Qiskit circuit version to be used for simulations etc.
+        """
+        if self._qiskit_circ is None:
+            self._qiskit_circ = build_qiskit_circ(self)
+
+        return self._qiskit_circ
+
+    def expressivity(self, config: ProxyConfig) -> float:
+        """
+        Expressivity of the circuit, following the definition from {THE PAPER}
+        """  # TODO: Link Paper
+        if self._expressivity is None:
+            self._expressivity = float(calculate_expressivity(self, config)[0])
+        return self._expressivity
+
+    def entanglement(self, config: ProxyConfig) -> float:
+        """
+        Entanglement of the circuit, following the definition from {THE PAPER}
+        """  # TODO: Link Paper
+        if self._entanglement is None:
+            self._entanglement = float(calculate_entanglement(self, config)[0])
+
+        return self._entanglement
+
+    def fidelity(self, config: ProxyConfig) -> float:
+        """
+        Approximated fidelity of the circuit
+        """
+        if self._fidelity is None:
+            self._fidelity = float(calculate_fidelity(self, config)[0])
+
+        return self._fidelity

src/quantum_circuit/proxies.py

@@ -0,0 +1,94 @@
+from itertools import pairwise
+from typing import TYPE_CHECKING
+
+import numpy as np
+from numpy.typing import NDArray
+from qiskit import transpile
+from qiskit_aer.backends.aer_simulator import AerSimulator
+from qiskit_aer.backends.aerbackend import AerBackend
+from scipy import stats
+from tqdm import tqdm
+
+if TYPE_CHECKING:
+    from quantum_circuit import ParametrizedQuantumCircuit
+    from quantum_circuit.proxy_config import ProxyConfig
+
+
+def single_circuit_param_fidelity(circ: ParametrizedQuantumCircuit, samples: int, backend: AerBackend) -> float:
+    qc, thetas = circ.circ
+    qc.save_statevector()
+
+    tqc = transpile(qc, backend)
+
+    number_of_initial_circuits = 2 * samples
+
+    params: NDArray[np.float64] = np.random.uniform(-np.pi, np.pi, (len(thetas), number_of_initial_circuits))
+
+    binds = [{param: params[idx] for idx, param in enumerate(thetas.params)}]
+
+    job = backend.run([tqc], parameter_binds=binds)
+    result = job.result()
+
+    sv = np.array(
+        [np.asarray(result.get_statevector(i), dtype=np.complex128) for i in range(number_of_initial_circuits)]
+    )
+    left = sv[:samples]
+    right = sv[samples:]
+
+    return np.power(np.absolute((left * right.conjugate()).sum(-1)), 2)
+
+
+def calculate_fidelity(
+    circs: list[ParametrizedQuantumCircuit] | ParametrizedQuantumCircuit, config: ProxyConfig
+) -> NDArray[np.float64]:
+    if isinstance(circs, ParametrizedQuantumCircuit):
+        circs = [circs]
+    raise NotImplementedError
+
+
+def calculate_expressivity(
+    circs: list[ParametrizedQuantumCircuit] | ParametrizedQuantumCircuit, config: ProxyConfig
+) -> NDArray[np.float64]:
+    tqdm_depth = 0
+    if isinstance(circs, ParametrizedQuantumCircuit):
+        circs = [circs]
+
+    qubits = config.qubits
+    samples = config.expressivity_samples
+    bin_count = config.expressivity_bins
+    force_recalculate = config.force_recalculate
+    bins: np.ndarray[tuple[int], np.dtype[np.float64]] = np.linspace(0.0, 1.0, bin_count + 1)
+
+    haar_power = (1 << qubits) - 1
+    lower_edges: np.ndarray[tuple[int], np.dtype[np.float64]] = -1.0 * np.power(1.0 - bins[:-1], haar_power)
+    higher_edges: np.ndarray[tuple[int], np.dtype[np.float64]] = -1.0 * np.power(1.0 - bins[1:], haar_power)
+    haar_values: np.ndarray[tuple[int], np.dtype[np.float64]] = higher_edges - lower_edges
+
+    backend = AerSimulator(method="statevector")
+
+    fidelities = np.zeros((len(circs)))
+    for idx, circuit in tqdm(enumerate(circs), position=tqdm_depth, desc="Computing Fidelities", leave=False):
+        if not force_recalculate and circs[idx]._expressivity is not None:
+            continue
+        fidelities[idx] = single_circuit_param_fidelity(circuit, samples, backend)
+
+    expressivity = np.zeros((len(circs)))
+    for idx, fid in tqdm(enumerate(fidelities), position=tqdm_depth, desc="Computing Expressibility"):
+        if not force_recalculate and circs[idx]._expressivity is not None:
+            continue
+        bin_idx = np.floor(fid * bin_count).astype(int)
+        num = np.array([len(bin_idx[bin_idx == i]) for i in range(bin_count)])
+
+        expressivity[idx] = -stats.entropy(num, haar_values)
+        # NOTE: I'm setting the expressivity here directly, for caching
+        circs[idx]._expressivity = float(expressivity[idx])
+
+    return expressivity
+
+
+def calculate_entanglement(
+    circs: list[ParametrizedQuantumCircuit] | ParametrizedQuantumCircuit, config: ProxyConfig
+) -> NDArray[np.float64]:
+    if isinstance(circs, ParametrizedQuantumCircuit):
+        circs = [circs]
+    raise NotImplementedError

src/quantum_circuit/proxy_config.py

@@ -0,0 +1,17 @@
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class ProxyConfig:
+    """
+    Configuration for various proxies, defaults are included for every setting so only the
+    non-default values need to be changed when creating one
+    """
+
+    qubits: int
+    force_recalculate: bool = False
+
+    entanglement_samples: int = 1000
+
+    expressivity_samples: int = 1000
+    expressivity_bins: int = 100

src/quantum_circuit/qiskit_helpers.py

@@ -0,0 +1,11 @@
+from typing import TYPE_CHECKING
+
+from qiskit import QuantumCircuit
+from qiskit.circuit.parametervector import ParameterVector
+
+if TYPE_CHECKING:
+    from quantum_circuit import ParametrizedQuantumCircuit
+
+
+def build_qiskit_circ(pqc: "ParametrizedQuantumCircuit") -> tuple[QuantumCircuit, ParameterVector]:
+    raise NotImplementedError

src/settings.py

@@ -0,0 +1,6 @@
+from dataclasses import dataclass
+
+
+@dataclass
+class QuantumArchitectureSearchSettings:
+    qubits: int

src/tf_qas/__init__.py

@@ -0,0 +1,8 @@
+from dataclasses import dataclass
+
+from settings import QuantumArchitectureSearchSettings
+
+
+@dataclass
+class TrainingFreeSettings(QuantumArchitectureSearchSettings):
+    sample_size: int