qadence 1.7.8__py3-none-any.whl → 1.9.0__py3-none-any.whl

qadence/measurements/shadow.py

@@ -1,40 +1,37 @@
 from __future__ import annotations
 
-from collections import Counter
-from functools import reduce
-
 import numpy as np
 import torch
 from torch import Tensor
 
 from qadence.backend import Backend
 from qadence.backends.pyqtorch import Backend as PyQBackend
-from qadence.blocks import AbstractBlock, chain, kron
-from qadence.blocks.block_to_tensor import HMAT, IMAT, SDAGMAT, ZMAT, block_to_tensor
+from qadence.blocks import AbstractBlock, KronBlock, chain, kron
+from qadence.blocks.block_to_tensor import HMAT, IMAT, SDAGMAT
 from qadence.blocks.composite import CompositeBlock
 from qadence.blocks.primitive import PrimitiveBlock
 from qadence.blocks.utils import get_pauli_blocks, unroll_block_with_scaling
 from qadence.circuit import QuantumCircuit
 from qadence.engines.differentiable_backend import DifferentiableBackend
-from qadence.noise import Noise
-from qadence.operations import X, Y, Z
+from qadence.measurements.utils import get_qubit_indices_for_op
+from qadence.noise import NoiseHandler
+from qadence.operations import H, I, SDagger, X, Y, Z
 from qadence.types import Endianness
-from qadence.utils import P0_MATRIX, P1_MATRIX
 
 pauli_gates = [X, Y, Z]
-
+pauli_rotations = [
+    lambda index: H(index),
+    lambda index: SDagger(index) * H(index),
+    lambda index: None,
+]
 
 UNITARY_TENSOR = [
-    ZMAT @ HMAT,
-    SDAGMAT.squeeze(dim=0) @ HMAT,
+    HMAT,
+    HMAT @ SDAGMAT,
     IMAT,
 ]
 
 
-def identity(n_qubits: int) -> Tensor:
-    return torch.eye(2**n_qubits, dtype=torch.complex128)
-
-
 def _max_observable_weight(observable: AbstractBlock) -> int:
     """
     Get the maximal weight for the given observable.
@@ -88,27 +85,40 @@ def number_of_samples(
     return N, K
 
 
-def local_shadow(sample: Counter, unitary_ids: list) -> Tensor:
+def nested_operator_indexing(
+    idx_array: np.ndarray,
+) -> list:
+    """Obtain the list of rotation operators from indices.
+
+    Args:
+        idx_array (np.ndarray): Indices for obtaining the operators.
+
+    Returns:
+        list: Map of rotations.
     """
-    Compute local shadow by inverting the quantum channel for each projector state.
+    if idx_array.ndim == 1:
+        return [pauli_rotations[int(ind_pauli)](i) for i, ind_pauli in enumerate(idx_array)]  # type: ignore[abstract]
+    return [nested_operator_indexing(sub_array) for sub_array in idx_array]
+
+
+def kron_if_non_empty(list_operations: list) -> KronBlock | None:
+    filtered_op: list = list(filter(None, list_operations))
+    return kron(*filtered_op) if len(filtered_op) > 0 else None
 
-    See https://arxiv.org/pdf/2002.08953.pdf
-    Supplementary Material 1 and Eqs. (S17,S44).
 
-    Expects a sample bitstring in ILO.
+def extract_operators(unitary_ids: np.ndarray, n_qubits: int) -> list:
+    """Sample `shadow_size` rotations of `n_qubits`.
+
+    Args:
+        unitary_ids (np.ndarray): Indices for obtaining the operators.
+        n_qubits (int): Number of qubits
+    Returns:
+        list: Pauli strings.
     """
-    bitstring = list(sample.keys())[0]
-    local_density_matrices = []
-    for bit, unitary_id in zip(bitstring, unitary_ids):
-        proj_mat = P0_MATRIX if bit == "0" else P1_MATRIX
-        unitary_tensor = UNITARY_TENSOR[unitary_id].squeeze(dim=0)
-        local_density_matrices.append(
-            3 * (unitary_tensor.adjoint() @ proj_mat @ unitary_tensor) - identity(1)
-        )
-    if len(local_density_matrices) == 1:
-        return local_density_matrices[0]
-    else:
-        return reduce(torch.kron, local_density_matrices)
+    operations = nested_operator_indexing(unitary_ids)
+    if n_qubits > 1:
+        operations = [kron_if_non_empty(ops) for ops in operations]
+    return operations
 
 
 def classical_shadow(
@@ -117,28 +127,23 @@ def classical_shadow(
     param_values: dict,
     state: Tensor | None = None,
     backend: Backend | DifferentiableBackend = PyQBackend(),
-    # FIXME: Changed below from Little to Big, double-check when Roland is back
-    noise: Noise | None = None,
+    noise: NoiseHandler | None = None,
     endianness: Endianness = Endianness.BIG,
-) -> list:
-    shadow: list = []
-    # TODO: Parallelize embarrassingly parallel loop.
-    for _ in range(shadow_size):
-        unitary_ids = np.random.randint(0, 3, size=(1, circuit.n_qubits))[0]
-        random_unitary = [
-            pauli_gates[unitary_ids[qubit]](qubit) for qubit in range(circuit.n_qubits)
-        ]
-        if len(random_unitary) == 1:
-            random_unitary_block = random_unitary[0]
+) -> tuple[np.ndarray, list[Tensor]]:
+    unitary_ids = np.random.randint(0, 3, size=(shadow_size, circuit.n_qubits))
+    shadow: list = list()
+    all_rotations = extract_operators(unitary_ids, circuit.n_qubits)
+
+    for i in range(shadow_size):
+        if all_rotations[i]:
+            rotated_circuit = QuantumCircuit(
+                circuit.register, chain(circuit.block, all_rotations[i])
+            )
         else:
-            random_unitary_block = kron(*random_unitary)
-        rotated_circuit = QuantumCircuit(
-            circuit.n_qubits,
-            chain(circuit.block, random_unitary_block),
-        )
+            rotated_circuit = circuit
         # Reverse endianness to get sample bitstrings in ILO.
         conv_circ = backend.circuit(rotated_circuit)
-        samples = backend.sample(
+        batch_samples = backend.sample(
             circuit=conv_circ,
             param_values=param_values,
             n_shots=1,
@@ -146,97 +151,61 @@ def classical_shadow(
             noise=noise,
             endianness=endianness,
         )
-        batched_shadow = []
-        for batch in samples:
-            batched_shadow.append(local_shadow(sample=batch, unitary_ids=unitary_ids))
-        shadow.append(batched_shadow)
-
-    # Reshape the shadow by batches of samples instead of samples of batches.
-    # FIXME: Improve performance.
-    return [list(s) for s in zip(*shadow)]
-
-
-def reconstruct_state(shadow: list) -> Tensor:
-    """Reconstruct the state density matrix for the given shadow."""
-    return reduce(torch.add, shadow) / len(shadow)
-
-
-def compute_traces(
-    qubit_support: tuple,
-    N: int,
-    K: int,
-    shadow: list,
-    observable: AbstractBlock,
-    endianness: Endianness = Endianness.BIG,
-) -> list:
-    floor = int(np.floor(N / K))
-    traces = []
-    # TODO: Parallelize embarrassingly parallel loop.
-    for k in range(K):
-        reconstructed_state = reconstruct_state(shadow=shadow[k * floor : (k + 1) * floor])
-        # Reshape the observable matrix to fit the density matrix dimensions
-        # by filling indentites.
-        # Please note the endianness is also flipped to get results in LE.
-        # FIXME: Changed below from Little to Big, double-check when Roland is back
-        # FIXME: Correct these comments.
-        trace = (
-            (
-                block_to_tensor(
-                    block=observable,
-                    qubit_support=qubit_support,
-                    endianness=Endianness.BIG,
-                ).squeeze(dim=0)
-                @ reconstructed_state
-            )
-            .trace()
-            .real
-        )
-        traces.append(trace)
-    return traces
+        shadow.append(batch_samples)
+    bitstrings = list()
+    batchsize = len(batch_samples)
+    for b in range(batchsize):
+        bitstrings.append([list(batch[b].keys())[0] for batch in shadow])
+    bitstrings_torch = [
+        1 - 2 * torch.stack([torch.tensor([int(b_i) for b_i in sample]) for sample in batch])
+        for batch in bitstrings
+    ]
+    return unitary_ids, bitstrings_torch
 
 
 def estimators(
-    qubit_support: tuple,
     N: int,
     K: int,
-    shadow: list,
+    unitary_shadow_ids: np.ndarray,
+    shadow_samples: Tensor,
     observable: AbstractBlock,
-    endianness: Endianness = Endianness.BIG,
 ) -> Tensor:
     """
-    Return estimators (traces of observable times mean density matrix).
-
-    for K equally-sized shadow partitions.
+    Return trace estimators from the samples for K equally-sized shadow partitions.
 
     See https://arxiv.org/pdf/2002.08953.pdf
     Algorithm 1.
     """
-    # If there is no Pauli-Z operator in the observable,
-    # the sample can't "hit" that measurement.
+
+    obs_qubit_support = observable.qubit_support
     if isinstance(observable, PrimitiveBlock):
-        if type(observable) == Z:
-            traces = compute_traces(
-                qubit_support=qubit_support,
-                N=N,
-                K=K,
-                shadow=shadow,
-                observable=observable,
-                endianness=endianness,
-            )
-        else:
-            traces = [torch.tensor(0.0)]
+        if isinstance(observable, I):
+            return torch.tensor(1.0, dtype=torch.get_default_dtype())
+        obs_to_pauli_index = [pauli_gates.index(type(observable))]
+
     elif isinstance(observable, CompositeBlock):
-        if Z in observable:
-            traces = compute_traces(
-                qubit_support=qubit_support,
-                N=N,
-                K=K,
-                shadow=shadow,
-                observable=observable,
-                endianness=endianness,
-            )
+        obs_to_pauli_index = [
+            pauli_gates.index(type(p)) for p in observable.blocks if not isinstance(p, I)  # type: ignore[arg-type]
+        ]
+        ind_I = set(get_qubit_indices_for_op((observable, 1.0), I(0)))
+        obs_qubit_support = tuple([ind for ind in observable.qubit_support if ind not in ind_I])
+
+    floor = int(np.floor(N / K))
+    traces = []
+    for k in range(K):
+        indices_match = np.all(
+            unitary_shadow_ids[k * floor : (k + 1) * floor, obs_qubit_support]
+            == obs_to_pauli_index,
+            axis=1,
+        )
+        if indices_match.sum() > 0:
+            trace = torch.prod(
+                shadow_samples[k * floor : (k + 1) * floor][indices_match][:, obs_qubit_support],
+                axis=-1,
+            ).sum() / sum(indices_match)
+            traces.append(trace)
         else:
-            traces = [torch.tensor(0.0)]
+            traces.append(torch.tensor(0.0))
     return torch.tensor(traces, dtype=torch.get_default_dtype())
 
 
@@ -249,7 +218,7 @@ def estimations(
     confidence: float = 0.1,
     state: Tensor | None = None,
     backend: Backend | DifferentiableBackend = PyQBackend(),
-    noise: Noise | None = None,
+    noise: NoiseHandler | None = None,
     endianness: Endianness = Endianness.BIG,
 ) -> Tensor:
     """Compute expectation values for all local observables using median of means."""
@@ -259,7 +228,7 @@ def estimations(
     N, K = number_of_samples(observables=observables, accuracy=accuracy, confidence=confidence)
     if shadow_size is not None:
         N = shadow_size
-    shadow = classical_shadow(
+    unitaries_ids, batch_shadow_samples = classical_shadow(
         shadow_size=N,
         circuit=circuit,
         param_values=param_values,
@@ -272,18 +241,17 @@ def estimations(
     for observable in observables:
         pauli_decomposition = unroll_block_with_scaling(observable)
         batch_estimations = []
-        for batch in shadow:
+        for batch in batch_shadow_samples:
             pauli_term_estimations = []
             for pauli_term in pauli_decomposition:
                 # Get the estimators for the current Pauli term.
                 # This is a tensor<float> of size K.
                 estimation = estimators(
-                    qubit_support=circuit.block.qubit_support,
                     N=N,
                     K=K,
-                    shadow=batch,
+                    unitary_shadow_ids=unitaries_ids,
+                    shadow_samples=batch,
                     observable=pauli_term[0],
-                    endianness=endianness,
                 )
                 # Compute the median of means for the current Pauli term.
                 # Weigh the median by the Pauli term scaling.
@@ -302,7 +270,7 @@ def compute_expectation(
     options: dict,
     state: Tensor | None = None,
     backend: Backend | DifferentiableBackend = PyQBackend(),
-    noise: Noise | None = None,
+    noise: NoiseHandler | None = None,
     endianness: Endianness = Endianness.BIG,
 ) -> Tensor:
     """

qadence/measurements/tomography.py

@@ -10,7 +10,7 @@ from qadence.blocks.utils import unroll_block_with_scaling
 from qadence.circuit import QuantumCircuit
 from qadence.engines.differentiable_backend import DifferentiableBackend
 from qadence.measurements.utils import iterate_pauli_decomposition
-from qadence.noise import Noise
+from qadence.noise import NoiseHandler
 from qadence.utils import Endianness
 
 
@@ -21,7 +21,7 @@ def compute_expectation(
     options: dict,
     state: Tensor | None = None,
     backend: Backend | DifferentiableBackend = PyQBackend(),
-    noise: Noise | None = None,
+    noise: NoiseHandler | None = None,
     endianness: Endianness = Endianness.BIG,
 ) -> Tensor:
     """Basic tomography protocol with rotations.

qadence/measurements/utils.py

@@ -14,8 +14,8 @@ from qadence.backends.pyqtorch import Backend as PyQBackend
 from qadence.blocks import AbstractBlock, PrimitiveBlock, chain
 from qadence.circuit import QuantumCircuit
 from qadence.engines.differentiable_backend import DifferentiableBackend
-from qadence.noise import Noise
-from qadence.operations import H, SDagger, X, Y, Z
+from qadence.noise import NoiseHandler
+from qadence.operations import H, I, SDagger, X, Y
 from qadence.parameters import evaluate
 from qadence.utils import Endianness
 
@@ -113,7 +113,7 @@ def rotate(circuit: QuantumCircuit, pauli_term: Tuple[AbstractBlock, Basic]) ->
 
     rotations = []
 
-    for op, gate in [(X(0), Z), (Y(0), SDagger)]:
+    for op, gate in [(X(0), I), (Y(0), SDagger)]:
         qubit_indices = get_qubit_indices_for_op(pauli_term, op=op)
         for index in qubit_indices:
             rotations.append(gate(index) * H(index))
@@ -128,7 +128,7 @@ def iterate_pauli_decomposition(
     n_shots: int,
     state: Tensor | None = None,
     backend: Backend | DifferentiableBackend = PyQBackend(),
-    noise: Noise | None = None,
+    noise: NoiseHandler | None = None,
     endianness: Endianness = Endianness.BIG,
 ) -> Tensor:
     """Estimate total expectation value by averaging all Pauli terms.

qadence/mitigations/analog_zne.py

@@ -15,7 +15,7 @@ from qadence.blocks.analog import ConstantAnalogRotation, InteractionBlock
 from qadence.circuit import QuantumCircuit
 from qadence.measurements import Measurements
 from qadence.mitigations import Mitigations
-from qadence.noise import Noise
+from qadence.noise import NoiseHandler
 from qadence.operations import AnalogRot
 from qadence.transpile import apply_fn_to_blocks
 from qadence.utils import Endianness
@@ -44,7 +44,7 @@ def pulse_experiment(
     circuit: QuantumCircuit,
     observable: list[AbstractBlock],
     param_values: dict[str, Tensor],
-    noise: Noise,
+    noise: NoiseHandler,
     stretches: Tensor,
     endianness: Endianness,
     state: Tensor | None = None,
@@ -116,11 +116,12 @@ def noise_level_experiment(
     circuit: QuantumCircuit,
     observable: list[AbstractBlock],
     param_values: dict[str, Tensor],
-    noise: Noise,
+    noise: NoiseHandler,
     endianness: Endianness,
     state: Tensor | None = None,
 ) -> Tensor:
-    noise_probs = noise.options.get("noise_probs")
+    protocol, options = noise.protocol[-1], noise.options[-1]
+    noise_probs = options.get("noise_probs")
     zne_datasets: list = []
     # Get noisy density matrices.
     conv_circuit = backend.circuit(circuit)
@@ -152,7 +153,7 @@ def analog_zne(
     param_values: dict[str, Tensor] = {},
     state: Tensor | None = None,
     measurement: Measurements | None = None,
-    noise: Noise | None = None,
+    noise: NoiseHandler | None = None,
     mitigation: Mitigations | None = None,
     endianness: Endianness = Endianness.BIG,
 ) -> Tensor:
@@ -162,7 +163,7 @@ def analog_zne(
     backend = cast(Backend, backend)
     noise_model = mitigation.options.get("noise_model", None)
     if noise_model is None:
-        KeyError(f"A noise model should be choosen from {Noise.list()}. Got {noise_model}.")
+        raise KeyError("A noise model should be specified.")
     stretches = mitigation.options.get("stretches", None)
     if stretches is not None:
         extrapolated_exp_values = pulse_experiment(
@@ -195,7 +196,7 @@ def mitigate(
     param_values: dict[str, Tensor] = {},
     state: Tensor | None = None,
     measurement: Measurements | None = None,
-    noise: Noise | None = None,
+    noise: NoiseHandler | None = None,
     mitigation: Mitigations | None = None,
     endianness: Endianness = Endianness.BIG,
 ) -> Tensor:

qadence/mitigations/protocols.py

@@ -5,7 +5,7 @@ from collections import Counter
 from dataclasses import dataclass
 from typing import Callable, cast
 
-from qadence.noise.protocols import Noise
+from qadence.noise.protocols import NoiseHandler
 
 PROTOCOL_TO_MODULE = {
     "readout": "qadence.mitigations.readout",
@@ -45,7 +45,7 @@ class Mitigations:
 
 
 def apply_mitigation(
-    noise: Noise, mitigation: Mitigations, samples: list[Counter]
+    noise: NoiseHandler, mitigation: Mitigations, samples: list[Counter]
 ) -> list[Counter]:
     """Apply mitigation to samples."""
     mitigation_fn = mitigation.get_mitigation_fn()

qadence/mitigations/readout.py

@@ -10,8 +10,9 @@ from numpy.linalg import inv, matrix_rank, pinv
 from scipy.linalg import norm
 from scipy.optimize import LinearConstraint, minimize
 
+from qadence.backends.pyqtorch.convert_ops import convert_readout_noise
 from qadence.mitigations.protocols import Mitigations
-from qadence.noise.protocols import Noise
+from qadence.noise.protocols import NoiseHandler
 from qadence.types import ReadOutOptimization
 
 
@@ -69,7 +70,7 @@ def matrix_inv(K: npt.NDArray) -> npt.NDArray:
 
 
 def mitigation_minimization(
-    noise: Noise,
+    noise: NoiseHandler,
     mitigation: Mitigations,
     samples: list[Counter],
 ) -> list[Counter]:
@@ -88,10 +89,18 @@ def mitigation_minimization(
     Returns:
         Mitigated counts computed by the algorithm
     """
-    noise_matrices = noise.options.get("noise_matrix", noise.options["confusion_matrices"])
-    optimization_type = mitigation.options.get("optimization_type", ReadOutOptimization.MLE)
+
     n_qubits = len(list(samples[0].keys())[0])
+    readout_noise = convert_readout_noise(n_qubits, noise)
+    if readout_noise is None:
+        raise ValueError("Specify a noise source of type NoiseProtocol.READOUT.")
     n_shots = sum(samples[0].values())
+    noise_matrices = readout_noise.confusion_matrix
+    if noise_matrices.numel() == 0:
+        readout_noise.create_noise_matrix(n_shots)
+        noise_matrices = readout_noise.confusion_matrix
+    optimization_type = mitigation.options.get("optimization_type", ReadOutOptimization.MLE)
+
     corrected_counters: list[Counter] = []
 
     if optimization_type == ReadOutOptimization.CONSTRAINED:
@@ -156,5 +165,5 @@ def mitigation_minimization(
     return corrected_counters
 
 
-def mitigate(noise: Noise, mitigation: Mitigations, samples: list[Counter]) -> list[Counter]:
+def mitigate(noise: NoiseHandler, mitigation: Mitigations, samples: list[Counter]) -> list[Counter]:
     return mitigation_minimization(noise=noise, mitigation=mitigation, samples=samples)

qadence/ml_tools/__init__.py

@@ -1,16 +1,14 @@
 from __future__ import annotations
 
-from .config import AnsatzConfig, Callback, FeatureMapConfig, TrainConfig
+from .callbacks.saveload import load_checkpoint, load_model, write_checkpoint
+from .config import AnsatzConfig, FeatureMapConfig, TrainConfig
 from .constructors import create_ansatz, create_fm_blocks, observable_from_config
 from .data import DictDataLoader, InfiniteTensorDataset, OptimizeResult, to_dataloader
 from .models import QNN
 from .optimize_step import optimize_step as default_optimize_step
 from .parameters import get_parameters, num_parameters, set_parameters
-from .printing import print_metrics, write_tensorboard
-from .saveload import load_checkpoint, load_model, write_checkpoint
 from .tensors import numpy_to_tensor, promote_to, promote_to_tensor
-from .train_grad import train as train_with_grad
-from .train_no_grad import train as train_gradient_free
+from .trainer import Trainer
 
 # Modules to be automatically added to the qadence namespace
 __all__ = [
@@ -24,8 +22,6 @@ __all__ = [
     "QNN",
     "TrainConfig",
     "OptimizeResult",
-    "Callback",
-    "train_with_grad",
-    "train_gradient_free",
+    "Trainer",
     "write_checkpoint",
 ]

qadence/ml_tools/callbacks/__init__.py

@@ -0,0 +1,30 @@
+from __future__ import annotations
+
+from .callback import (
+    Callback,
+    LoadCheckpoint,
+    LogHyperparameters,
+    LogModelTracker,
+    PlotMetrics,
+    PrintMetrics,
+    SaveBestCheckpoint,
+    SaveCheckpoint,
+    WriteMetrics,
+)
+from .callbackmanager import CallbacksManager
+from .writer_registry import get_writer
+
+# Modules to be automatically added to the qadence.ml_tools.callbacks namespace
+__all__ = [
+    "CallbacksManager",
+    "Callback",
+    "LoadCheckpoint",
+    "LogHyperparameters",
+    "LogModelTracker",
+    "PlotMetrics",
+    "PrintMetrics",
+    "SaveBestCheckpoint",
+    "SaveCheckpoint",
+    "WriteMetrics",
+    "get_writer",
+]