QClassify is a Python framework for implementing variational quantum classifiers. The goal is to provide a generally customizable way of performing classification tasks using gate-model quantum devices. The quantum devices can be either simulated by a quantum simulator or a cloud-based quantum processor accessible via Rigetti Computing's Quantum Cloud Services.
Variational quantum classification is a paradigm of supervised quantum machine learning that has been investigated actively in the quantum computing community (See for instance Farhi and Neven, Schuld et al., Mitarai et al. and Havlicek et al.). The general framework adopted in the design of QClassify follows from these contributions in the literature. The workflow can be summarized in Figure 1 below.
Figure 1: Diagram illustrating the workflow of QClassify. Each rectangle represents a data object and each oval represents a method.
The main data structure describing the quantum classifier setting is in qclassifier.py. The implementation allows for modular design of the following components of a quantum classifier (Figure 1):
-
Encoder: transforms a classical data vector into a quantum state. See
encoder.py.-
1.1 Classical preprocessor: maps an input data vector to circuit parameters. See
preprocessing.py. -
1.2 Quantum state preparation: applies the parametrized circuit to an all-zero input state to generate a quantum state encoding the input data. See
encoding_circ.py.
-
-
Processor: extracts classical information from the encoded quantum state. See
processor.py.-
2.1 Quantum state transformation: applies a parametrized circuit to the encoded quantum state to transform it into a form more amenable for information extraction by measurement and classical postprocessing. See
proc_circ.py. -
2.2 Information extraction: extract classical information from the output quantum state. See
postprocessing.py.-
2.2.1 Measurement: repeatedly run the quantum circuit, perform measurements and collect measurement statistics
-
2.2.2 Classical postprocessing: Glean information from the measurement statistics and produce the output label of the quantum classifier.
-
-
To install QClassify using pip:
pip install qclassify
Try executing import qclassify to test the installation in your terminal.
To instead install QClassify from source, clone this repository, cd into it, and run:
git clone https://github.com/zapatacomputing/QClassify
cd QClassify
python -m pip install -e .
Note that the pyQuil version used requires Python 3.6 or later. For installation on a user QMI, please click here.
We provide a Jupyter notebook to demonstrate the utility of QClassify.
| Notebook | Feature(s) |
|---|---|
| qclassify_demo.ipynb | Uses a simple two-qubit circuit to learn the XOR dataset. |
We note that there is a lot of room for improvement and fixes. Please feel free to submit issues and/or pull requests!
When using QClassify for research projects, please cite:
Sukin Sim, Yudong Cao, Jonathan Romero, Peter D. Johnson and Alán Aspuru-Guzik. A framework for algorithm deployment on cloud-based quantum computers. arXiv:1810.10576. 2018.
Yudong Cao, Sukin (Hannah) Sim (Harvard)