vc.models module

This module is used to define quantum models.

To add a new model, you should implement the QModel interface and update get_model().

class vc.models.ExpectedValuesModel(backend, n_classes, n_layers=2, n_trainable_sublayers=2, scaling=0.5)

Bases: QModel

Model that uses angle encoding and expected values.

__init__(backend, n_classes, n_layers=2, n_trainable_sublayers=2, scaling=0.5)

Init ExpectedValuesModel.

This model implements a unitary \(U(x)\) of the form:

\[U(x) = W_{L + 1} \cdot S(x) \cdot W_{L} \cdot \ldots \cdot W_{2} \cdot S(x) \cdot W_{1}\]

where:

• \(x\) is the data to encode,

• \(S\) is an encoding circuit,

• \(W\) is a trainable circuit.

Parameters:

• backend (Dict[str, Union[str, Any]]) – see docstring of QModel.

• n_classes (int) – see docstring of QModel.

• n_layers (int) – number of layers in \(U(x)\) (equal to \(L\)).

• n_trainable_sublayers (int) – number of layers for each \(W\).

• scaling (float) – scaling to apply to the preprocessed data, see preprocess().

get_init_weights(random, random_state)

Get init weights between 0 and \(2\pi\).

Return type:

ndarray[Any, dtype[float64]]

get_qfunc()

Define callable based on circuit.

Return type:

Callable[[TensorType, TensorType], TensorType]

preprocess(X, min_max)

Maps input X in the range min_max to \((-\pi, \pi]\).

Parameters:

• X (ArrayLike) – input data with shape (n_samples, n_features).

• min_max (Tuple[float, float]) – minimum value and maximum value.

Return type:

ndarray[Any, dtype[float64]]

exception vc.models.ModelError(message)

Bases: Exception

Module exception.

__init__(message)

Init ModelError.

class vc.models.ModuloModel(backend, n_classes, n_layers=2, n_trainable_sublayers=2, scaling=0.5)

Bases: ProbabilitiesModel

Model that uses post-processing with modulo.

See docstring of vc.models.ProbabilitiesModel.

The post-processing strategy assigns a class to an \(n\)-bit string \(b\) according to the following formula:

\[f(b) = \left[b\right]_{10} \mod M\]

where:

• \(M\) is the number of classes,

• \([\cdot]_{10}\) is the decimal representation of the argument.

class vc.models.ParityModel(backend, n_classes, n_layers=2, n_trainable_sublayers=2, scaling=0.5)

Bases: ProbabilitiesModel

Model that uses parity post-processing.

See docstring of vc.models.ProbabilitiesModel.

The post-processing strategy assigns a class to an \(n\)-bit string \(b\) according to the following formula:

\[f(b) = \left[b_0 ... b_{m-2}\left(\bigoplus_{i=m-1}^{n-1} b_i\right) \right]_{10}\]

where:

• \(m=\lceil \log_2(M) \rceil\) with \(M\) being the number of classes,

• \(n\) is the number of bits,

• \([\cdot]_{10}\) is the decimal representation of the argument.

Reference: “Quantum Policy Gradient Algorithm with Optimized Action Decoding” by Meyer et al.

class vc.models.ProbabilitiesModel(backend, n_classes, n_layers=2, n_trainable_sublayers=2, scaling=0.5)

Bases: QModel, ABC

Generic model that uses probabilities.

__init__(backend, n_classes, n_layers=2, n_trainable_sublayers=2, scaling=0.5)

Init ProbabilitiesModel.

This model implements a unitary \(U(x)\) of the form:

\[U(x) = W_{L + 1} \cdot S(x) \cdot W_{L} \cdot \ldots \cdot W_{2} \cdot S(x) \cdot W_{1}\]

where:

• \(x\) is the data to encode,

• \(S\) is an encoding circuit,

• \(W\) is a trainable circuit.

Parameters:

• backend (Dict[str, Any]) – see docstring of QModel.

• n_classes (int) – see docstring of QModel.

• n_layers (int) – number of layers in \(U(x)\) (equal to \(L\)).

• n_trainable_sublayers (int) – number of layers for each \(W\).

• scaling (float) – scaling to apply to the preprocessed data, see preprocess().

get_init_weights(random, random_state)

Get init weights between \(0\) and \(2\pi\).

Return type:

ndarray[Any, dtype[float64]]

get_qfunc()

Get callable based on circuit.

Return type:

Callable[[TensorType, TensorType], TensorType]

preprocess(X, min_max)

Maps input X in the range min_max to \((-\pi, \pi]\).

Parameters:

• X (ArrayLike) – input data with shape (n_samples, n_features).

• min_max (Tuple[float, float]) – minimum value and maximum value.

Return type:

ndarray[Any, dtype[float64]]

class vc.models.QModel(backend, n_classes)

Bases: ABC

Abstract base class for quantum models.

__init__(backend, n_classes)

Init QModel.

Parameters:

• backend (Dict[str, Any]) – dictionary of the form {"name": str, "options": dict}, where the value for "name" is the name of a PennyLane device and the value for "options" is a dict to be passed as kwargs to PennyLane when creating the device. Example: {"name": "default.qubit", "options": {}}

• n_classes (int) – number of classes.

abstract get_init_weights(random, random_state)

Generate weights to be used as initial trainable parameters.

Return type:

ndarray[Any, dtype[float64]]

abstract get_qfunc()

Generate and return a quantum function.

Return type:

Callable[[TensorType, TensorType], TensorType]

abstract preprocess(X, min_max)

Convert X to features. This function should set self.n_qubits.

Return type:

ndarray[Any, dtype[float64]]

vc.models.get_model(model, backend, n_classes)

Create instance of a quantum model.

Parameters:

• model (Dict[str, Any]) – dictionary of the form {"name": str, "options": dict}. The value for "name" is used to determine the model class. The value for "options" should be a dict and is passed as kwargs to the constructor of the model class. Note: if there’s a "backend" key in the value for "options", it will be ignored.

• backend (Dict[str, Any]) – see docstring of QModel.

• n_classes (int) – number of classes.

Return type:

QModel

Returns:

An instance of a quantum model.