Quantum Reservoir Models¶

Quantum reservoir models use a fixed quantum circuit as a nonlinear feature map. The quantum circuit is not trained; a classical readout is fitted on the measured expectation-value features.

This keeps training cheap and makes the estimators useful for small tabular datasets, time-window features, and dynamical-system summaries.

Feature Map¶

QuantumReservoirFeatures maps each input row to Pauli-Z expectations:

from qml import QuantumReservoirFeatures

reservoir = QuantumReservoirFeatures(
    n_qubits=4,
    n_layers=2,
    seed=123,
)
features = reservoir.fit_transform(x_train)

Here x_train is the training feature matrix with one sample per row, and features is the transformed matrix of reservoir expectation-value features.

The circuit:

repeats or truncates each input row to the reservoir qubit count
angle-encodes bounded features
applies fixed random rotations
applies nearest-neighbor entanglers
returns one expectation feature per qubit

Constructor parameters:

Parameter	Meaning
`n_qubits`	Reservoir width. Defaults to input feature count.
`n_layers`	Number of fixed random reservoir layers.
`seed`	Random seed for fixed circuit weights.
`shots`	Optional finite-shot execution. `None` is analytic.
`input_scale`	Scale applied to bounded input angles.
`weight_scale`	Scale of fixed random circuit weights.

QuantumReservoirRegressor¶

QuantumReservoirRegressor fits a ridge readout on reservoir features.

from qml import QuantumReservoirFeatures, QuantumReservoirRegressor

model = QuantumReservoirRegressor(
    QuantumReservoirFeatures(n_qubits=4, n_layers=2, seed=123),
    alpha=1e-3,
)
model.fit(x_train, y_train)
pred = model.predict(x_test)

Here y_train is the continuous training-target vector, x_test is the test feature matrix, and pred contains predicted continuous target values.

The regressor exposes score(x, y) as negative mean-squared error and mean_absolute_error(x, y) as a convenience metric.

QuantumReservoirClassifier¶

QuantumReservoirClassifier fits a logistic readout on reservoir features.

from qml import QuantumReservoirClassifier, QuantumReservoirFeatures

model = QuantumReservoirClassifier(
    QuantumReservoirFeatures(n_qubits=4, n_layers=2, seed=123),
    c=1.0,
)
model.fit(x_train, y_train)
pred = model.predict(x_test)
proba = model.predict_proba(x_test)

Here y_train contains class labels, pred contains predicted class labels, and proba contains predicted class probabilities.

The classifier supports binary and multiclass labels through scikit-learn's logistic regression implementation.

When To Use¶

Reservoir models are useful when:

the input already contains short windows or summary features
a nonlinear representation is needed
full variational circuit training would be too expensive
repeatable fixed-feature baselines are useful

They are general QML estimators. Domain-specific simulators and feature extractors should stay in notebooks or user code.