polyadicqml

polyadicqml.circuitML

class polyadicqml.circuitML(make_circuit, nbqbits, nbparams, cbuilder)

Abstract Quantum ML circuit interface. Provides a unified interface to run multiple parametric circuits with different input and model parameters, agnostic of the backend, implemented in the subclasses.

Parameters

  • make_circuit (callable of signature self.make_circuit) – Function to generate the circuit corresponding to input x and params.

  • nbqbits (int) – Number of qubits.

  • nbparams (int) – Number of parameters.

  • cbuilder (circuitBuilder) – Circuit builder class to be used. It must correspond to the subclass implementation.

nbqbits

Number of qubits.

Type

int

nbparams

Number of parameters.

Type

int

grad(X, params, v=None, eps=None, nbshots=None, job_size=None)

Compute the gradient of the circuit w.r.t. parameters params on input X.

Uses finite differences of the circuit runs.

Parameters

  • X (array-like) – Input matrix of shape (nb_samples, nb_features).

  • params (vector-like) – Parameter vector of length nb_params.

  • v (array-like) – Vector or matrix to right multiply the Jacobian with.

  • eps (float, optional) – Epsilon for finite differences. By default uses 1e-8 if nbshots is not provided, else uses \pi / \sqrt{\text{nbshots}}

  • nbshots (int, optional) – Number of shots for the circuit run, by default None. If None, uses the backend default.

  • job_size (int, optional) – Maximum job size, to split the circuit runs, by default None. If None, put all nb_samples in the same job.

Returns

Jacobian matix as an array of shape (nb_params, 2**nbqbits) if v is None, else Jacobian-vector product: J(circuit) @ v

Return type

array

make_circuit(bdr, x, params)

Generate the circuit corresponding to input x and params. NOTE: This function is to be provided by the user, with the present signature.

Parameters

  • bdr (circuitBuilder) – A circuit builder.

  • x (vector-like) – Input sample

  • params (vector-like) – Parameter vector.

Returns

Instructed builder

Return type

circuitBuilder

random_params(seed=None)

Generate a valid vector of random parameters.

Parameters

seed (int, optional) – random seed, by default None

Returns

Vector of random parameters.

Return type

vector

run(X, params, nbshots=None, job_size=None)

Run the circuit with input X and parameters params.

Parameters

  • X (array-like) – Input matrix of shape (nb_samples, nb_features).

  • params (vector-like) – Parameter vector.

  • nbshots (int, optional) – Number of shots for the circuit run, by default None. If None, uses the backend default.

  • job_size (int, optional) – Maximum job size, to split the circuit runs, by default None. If None, put all nb_samples in the same job.

Returns

Bitstring counts as an array of shape (nb_samples, 2**nbqbits)

Return type

array

polyadicqml.Classifier

class polyadicqml.Classifier(circuit, bitstr, **kwargs)

Class for quantum classifiers. Defines the API using the scikit-learn format.

Parameters

  • circuit (circuitML) – Quantum circuit to simulate, how to use and store is defined in child classes.

  • bitstr (list of int or list of str) – Which bitstrings should correspond to each class. The number of classes for the classification is defined by the number of elements.

  • params (vector, optional) – Initial model paramters. If None (default) uses circuitML.random_params.

  • nbshots (int, optional) – Number of shots for the quantum circuit. If 0, negative or None, then exact proabilities are computed, by default None.

  • nbshots_increment (float, int or callable, optional) – How to increase the number of shots as optimization progress. If float or int, the increment arise every nbshots_incr_delay iterations: if float, then the increment is multiplicative; if int, then it is added. If callable, the new nbshots is computed by calling nbshots_increment(nbshots, n_iter, loss_value).

  • nbshots_incr_delay (int, optional) – After how many iteration nb_shots has to increse. By default 20, if nbshots_increment is given

  • loss (callable, optional) – Loss function, by default Negative LogLoss (Cross entropy).

  • job_size (int, optional) – Number of runs for each circuit job, by default the number of observations.

  • budget (int, optional) – Maximum number of optimization steps, by default 100

  • name (srt, optional) – Name to identify this classifier.

  • save_path (str, optional) – Where to save intermediate training results, by deafult None. If None, intermediate results are not saved.

bitstr

Bitstrings (as int) on which to read the classes

Type

list[int]

nbshots

Number of shots to run circuit

Type

int

job_size

Number of circuits to run in each backend job

Type

int

nfev

Number if times the circuit has been run

Type

int

fit(input_train, target_train, batch_size=None, **kwargs)

Fit the model according to the given training data.

Parameters

  • input_train (array) – Training design matrix.

  • target_train (vector) – Labels corresponding to input_train.

  • batch_size (int, optional) – Minibatches size, by default None. If none uses the full dataset with rndom shuffle at each iteration.

  • method (str, optional) – Optimization method, by default BFGS

  • bounds (sequence, optional) – Bounds on variables for L-BFGS-B, TNC, SLSQP, Powell, and trust-constr methods as a sequence of (min, max) pairs for each element in x. None is used to specify no bound.

  • options (dict, optional) – Optimizer options, by default {‘maxiter’: budget}

  • save_loss_progress (bool, optional) – Whether to store the loss progress, by default False

  • save_output_progress (file path, optional) – Path where to save the output evolution , by default None. If none is given, the output is not saved.

  • seed (int, optional) – Random seed, by default None

Returns

self

Return type

Classifier

info_dict()

Returns a dictionary containing models information.

Returns

Information dictionary

Return type

dict

predict(X)

Compute the predicted class for each input point of the design matrix.

Parameters

X (array) – Design matrix of n samples

Returns

Labels vector

Return type

vector

predict_proba(X, params=None)

Compute the bitstring probabilities associated to each input point of the design matrix.

Parameters

  • X (array) – Design matrix of n samples

  • params (vector, optional) – Circuit parameters, by default None. If not given, model parameters are used.

Returns

Predicted bitstring probabilities. Rows correspond to samples and columns to bitstrings, whose order is defined in bitstr.

Return type

array

proba_to_label(proba) → numpy.ndarray

Transforms a matrix of real values in integer labels.

Parameters

proba (array) – Real valued array

Returns

Labels vector

Return type

vector

run_circuit(X, params=None)

Run the circuit with input X and parameters params.

Parameters

  • X (array-like) – Input matrix of shape (nb_samples, nb_features).

  • params (vector-like, optional) – Parameter vector, by default uses the model params

Returns

Bitstring counts as an array of shape (nb_samples, 2**nbqbits)

Return type

array

set_bitstr(bitstr)

Bitstring setter

Parameters

bitstr (list[str] or list[int]) – Bitstrings on which to read the class predictions.

Raises

TypeError – If bitstrings are of wrong type or have eterogenous types

set_circuit(circuit)

Set the circuit after testing for validity.

For a circuit to be valid, it has to be an instance of circuitML and, in case self already has a circuit, to use the same make_circuit function.

Parameters

circuit (circuitML) – QML circuit

Raises

Union[TypeError, ValueError] – If the circuit is invalid.

set_loss(loss=None)

Loss function setter.

Parameters

loss (callable, optional) – Loss function of the form loss(y_true, y_pred, labels), by default None. If None is given, nothing happens.

set_params(params)

Parameters setter

Parameters

params (vector) – Parameters vector

polyadicqml.circuitBuilder

class polyadicqml.circuitBuilder(nbqbits, *args, **kwargs)

Builder to create parametrized circuit with repetitive structures, by defining general operations without directly writing gates.

Parameters

nbqbits (int) – Number of qubits.

alldiam(idx=None)

Add X-rotation of \pi/2.

Parameters

idx (iterable, optional) – Indices on which to apply the rotation, by default None. If None, apply to all qubits.

Returns

self

Return type

circuitBuilder

Raises

  • ValueError – If idx is out of range

  • TypeError – If the indices are not int

allin(theta)

Add input gate to all qubits and input theta.

Parameters

theta (list-like) – Parameters to input.

Returns

self

Return type

circuitBuilder

circuit()

Return the built circuit.

Raises

NotImplementedError –

cz(a, b)

Add CZ gate between qubits a and b

Parameters

  • a (int) – Control qubit

  • b (int) – Target qubit.

Returns

self

Return type

circuitBuilder

Raises

  • ValueError – If a or b are out of range

  • TypeError – If the indices are not int

input(idx, theta)

Add input gate.

It correspond to a rotation of RX(\pi/2) RZ(theta) RX(\pi/2).

Parameters

  • idx (Union[iterable, int]) – Index[-ices] of qubits on which to input theta.

  • theta (Union[list-like, float]) – Parameter[s] to input. Has to have the same length as idx.

Returns

self

Return type

circuitBuilder

Raises

  • ValueError – If idx is out of range

  • TypeError – If the indices are not int

measure_all()

Add measurement.

Returns

self

Return type

circuitBuilder