QRPCA

qrpca works similarly to sklean.decomposition, but employs a QR-based PCA decomposition and supports CUDA acceleration via torch.

How to install qrpca

The qrpca can be installed by the PyPI and pip:

pip install qrpca

If you download the repository, you can also install it in the qrpca directory:

git clone https://github.com/xuquanfeng/qrpca
cd qrpca
python setup.py install

You can access it by clicking on Github-qrpca .

Usage

Here is a demo for the use of qrpca.

The following are the results of retaining principal components containing 95% of the information content by principal component analysis.

You can set the parameter n_components to a value between 0 and 1 to execute the PCA on the corresponding proportion of the entire data, or set it to an integer number to reserve the n_omponents components.

import torch
import numpy as np
from qrpca.decomposition import qrpca
from qrpca.decomposition import svdpca

# Generate the random data
demo_data = torch.rand(60000,2000)
n_com = 0.95

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
# qrpca
pca = qrpca(n_component_ratio=n_com,device=device) # The percentage of information retained.
# pca = qrpca(n_component_ratio=10,device=device) # n principal components are reserved.
demo_qrpca = pca.fit_transform(demo_data)
print(demo_pca)

# SVDPCA
pca = svdpca(n_component_ratio=n_com,device=device)
demo_svdpca = pca.fit_transform(demo_data)
print(demo_svdpca)

Comparision with sklearn

The methods and usage of qrpca are almost identical to those of sklearn.decomposition.PCA. If you want to switch from sklearn to qrpca, all you have to do is change the import and declare the device if you have a GPU, and that's it.

And here's an illustration of how minimal the change is when different PCA is used:

• qrpca.decomposition.qrpca

from qrpca.decomposition import qrpca    
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
pca = qrpca(n_component_ratio=n_com,device=device)
demo_qrpca = pca.fit_transform(demo_data)

• qrpca.decomposition.svdpca

from qrpca.decomposition import svdpca
pca = svdpca(n_component_ratio=n_com)
demo_svdpca = pca.fit_transform(demo_data)

• sklearn.decomposition.PCA

from sklearn.decomposition import PCA
pca = PCA(n_components=n_com)
demo_pca = pca.fit_transform(demo_data)

Performance benchmark sklearn

With the acceleration of GPU computation, the speed of both QR decomposition and singular value decomposition in qrpca is much higher than that in sklearn

We run the different PCA methods on data with different numbers of rows and columns, and then we compare their PCA degradation times and plotted the distribution of the times. Here are the two plots.

Comparison of PCA degradation time with different number of rows and different methods for the case of 1000 columns.

Comparison of PCA reduction time with different number of columns and different methods for the case of 30000 rows.

We can see from the above two facts that qrpca may considerably cut program run time by using GPU acceleration, while also having a very cheap migration cost and a guaranteed impact.

Requirements

• numpy>=1.21.1
• pandas>=1.3.5
• torch>=1.8.1
• torchvision>=0.8.0
• cudatoolkit>=0.7.1
• scikit-learn>=1.0.2

Copyright & License

2022 Xu Quanfeng (xuquanfeng@shao.ac.cn) & Rafael S. de Souza (drsouza@shao.ac.cn) & Shen Shiyin (ssy@shao.ac.cn) & Peng Chen (pengchzn@gmail.com)

This program is free software: you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.

References

• Sharma, Alok and Paliwal, Kuldip K. and Imoto, Seiya and Miyano, Satoru 2013, International Journal of Machine Learning and Cybernetics, 4, 6, doi: 10.1007/s13042-012-0131-7 .

Citing qrpca

If you want to cite qrpca, please use the following citations.

Software Citation: Xu Quanfeng, & Rafael S. de Souza. (2022). PCA algorithm of QR accelerated SVD decomposition (1.5). Zenodo. https://doi.org/10.5281/zenodo.6555926