Qreservoir

Qreservoir is a lightweight python package built on top of qulacs to simulate quantum extreme learning and quantum reservoir computing models.

Qreservoir is licensed under the MIT license.

Tutorial and API documents

Documentation and API: https://owenagnel.github.io/qreservoir.

Quick Install for Python

pip install qreservoir

Uninstall Qreservoir:

pip uninstall qreservoir

Features

Fast simulation of quantum extreme learning machine and quantum reservoir computing.

Python sample code

from qulacs import Observable
from sklearn.linear_model import LinearRegression

from qreservoir.datasets import Complex_Fourrier
from qreservoir.encoders import ExpEncoder
from qreservoir.models import QELModel
from qreservoir.reservoirs import RotationReservoir

dataset = Complex_Fourrier(complexity=1, size=1000, noise=0.0)

encoder = ExpEncoder(1, 1, 3) # 1 feature, 1 layer, 1 qubit per feature
reservoir = RotationReservoir(encoder, 0, 10)  # 0 ancilla qubits, 10 depth

observables = [Observable(3) for _ in range(9)] # create observable set
for i, ob in enumerate(observables[:3]):
    ob.add_operator(1.0, f"X {i}")
for i, ob in enumerate(observables[3:6]):
    ob.add_operator(1.0, f"Z {i}")
for i, ob in enumerate(observables[6:]):
    ob.add_operator(1.0, f"Y {i}")

model = QELModel(reservoir, observables, LinearRegression()) # observable is a qulacs Observable object
X, _, y, _ = dataset.get_train_test()
model.fit(X, y)
print(model.score(X, y))

How to cite

Please cite the original qulacs arXiv paper: Qulacs: a fast and versatile quantum circuit simulator for research purpose

Future improvements

• Model creation currently a bit clunky. Models should take encoders and reservoirs as two seperate arguments and size of inputs should be determined dynamically when fit/score are called. This is made more difficult by the fact we wish to extract variance and concentration data from the models. Ideally we want RCModel and QELModel to implement scikit-learn's BaseEstimator interface.

• Additional tests should be written for datasets and new reservoirs/encoders

• Improve package structure by using python specific object oriented features.