This repository contains the code to reproduce the results in the paper An efficient quantum algorithm for ensemble classification using bagging. All the quantum implementations use the IBM qiskit package.
We propose a quantum algorithm to perform ensemble classification using bagging strategy. The underpinning idea is to generate several quantum trajectories in superposition to obtain B different transformations of the training set in only log(B) operations. This implies an exponential speedup in the size of the ensemble with respect to classical methods. Furthermore, when considering the overall time cost of the algorithm the training of a single weak classifier impacts additively rather than multiplicatively, as it usually happens.
To demonstrate this, we define simple routine of classification named Cosine Classifier and implement an ensemble with it. In particular, we conduct experiments on simulated data to show that (i) one execution of the quantum cosine classifier allows retrieving ensemble prediction, and (ii) the ensemble outperforms any of the single classifiers.
The code is organised in two parts:
-
The IPython notebooks cover point (i) implementing all the experiments in Section 4.1 and 4.2. Also, they cover all the technical details about the quantum cosine classifier and the generation of multiple independent trajectories in superposition,
-
The python scripts quantum_cosine_classifier.py and quantum_ensemble, instead, cover point (ii) and implement the experiments of Section 4.3 where it is shown that the quantum ensemble always outperforms the single classifier and has lower variance
The script Utils.py is used to import the needed packages and all of the custom routines to generate and plot data.
The script modeling.py contains the custom routines to define and execute the quantum algorithms.
The script Post_processing.py plots the results of quantum implementation.
For any issues or questions related to the code, open a new git issue or send a mail to antonio.macaluso@dfki.de