BI商业智能>> USPEC_USENC>> 返回
项目作者： huangdonghere

项目描述 ：
TKDE 2020: Ultra-Scalable Spectral Clustering and Ensemble Clustering (U-SPEC & U-SENC) #large-scale spectral clustering# #large-scale ensemble clustering#
高级语言： MATLAB
项目地址: git://github.com/huangdonghere/USPEC_USENC.git
创建时间： 2019-10-04T05:10:51Z
项目社区：https://github.com/huangdonghere/USPEC_USENC
开源协议：
下载


Ultra-Scalable Spectral Clustering and Ensemble Clustering

Overview

This repository provides the Matlab source code for two large-scale clustering algorithms, namely, Ultra-Scalable Spectral Clustering (U-SPEC) and Ultra-Scalable Ensemble Clustering (U-SENC), both of which have nearly linear time and space complexity and are capable of robustly and efficiently partitioning ten-million-level nonlinearly-separable datasets on a PC with 64GB memory.

If you find this repository helpful for your research, please cite the paper below. 

  1. Dong Huang, Chang-Dong Wang, Jian-Sheng Wu, Jianhuang Lai, and Chee-Keong Kwoh.
  2. Ultra-Scalable Spectral Clustering and Ensemble Clustering, 
  3. IEEE Transactions on Knowledge and Data Engineering (TKDE), 2020, vol.32, no.6, pp.1212-1226. 
  4. DOI: https://doi.org/10.1109/TKDE.2019.2903410

Description of Files

Code

Function Description
demo_1_USPEC.m A demo of the U-SPEC algorithm.
demo_2_USENC.m A demo of the U-SENC algorithm.
USPEC.m Call this function to perform the U-SPEC algorithm.
USENC.m Call this function to perform the U-SENC algorithm.
litekmeans.m A fast implementation of k-means.
computeNMI.m Call this function to compute the NMI score.
synthesizeLargescaleDatasets.p Call this function to synthesize the five large-scale datasets, whose sizes range from one million to twenty million.
synthesizeLargescaleDatasets_withArbitrarySizes.p Produce the five synthetic datasets with arbitrary sizes.

Data

In this repository, we provide the files of the five real-world datasets, namely, PenDigits, USPS, Letters, MNIST, and Covertype. We also provide the MATLAB code to reproduce the five large-scale synthetic datasets used in our paper.

How to Reproduce the Synthetic Datasets?

  • To generate the five large-scale synthetic datasets, you can call the synthesizeLargescaleDatasets function, which has just one input parameter. Note that this input parameter can be set to one of the five data names:

    • ‘TB1M’
    • ‘SF2M’
    • ‘CC5M’
    • ‘CG10M’
    • ‘Flower20M’

Example (to synthesize the CC5M dataset):

  1. ```
  2. synthesizeLargescaleDatasets('CC5M');
  3. % The synthesized dataset will be saved in 'data_CC5M.mat'.
  4. ```

  • To generate the five synthetic datasets with arbitrary sizes, you can call the synthesizeLargescaleDatasets_withArbitrarySizes function, which has two input parameters, that is

    • dataName: can be one of the five names: ‘TB’, ‘SF’, ‘CC’, ‘CG’, ‘Flower’
    • dataSize: can be set to any integers, provided that you have enough space to save the data.

Example (to synthesize a CG dataset with one million points):

  1. ```
  2. dataName = 'CG';
  3. dataSize = 1000000;
  4. synthesizeLargescaleDatasets_withArbitrarySizes(dataName, dataSize); 
  5. % The synthesized dataset will be saved in 'data_CG_1000000.mat'.
  6. ```

Further Questions?

Don’t hesitate to contact me if you have any questions regarding this work.
Email: huangdonghere at gmail dot com
Website: https://www.researchgate.net/publication/330760669