Purpose

This repository contains Matlab codes and clustering benchmark dataset.

This code provides the specific implementation of LoRD and B-LoRD, and reproduces some experimental results reported in the paper.

Steps

Both demo_train.m and demo_synthenic.m can be run directly.

You can select alg = "LoRD" or alg = "BLoRD" in Global settings to run the differernt models.

To change the dataset in demo_train.m, you just change the input of loadData() function as "ORL", "CHART", "USPS-1000", etc.

In demo_synthenic.m, you may set npi = [20, 40, 60, 80] to other value, such as npi = [10, 30, 70, 100].

Contents

The files in this repository are:

• demo_train.m: The code for reproduces the training results in the paper.
• demo_synthenic.m: The code for reproduces the synthenic experimen results in the paper.
• demo_DCD.m: The code for reproduces the DCD method.
• LoRD.m: The code for the proposed LoRD method.
• BLoRD.m: The code for the proposed B-LoRD method.
• Dykstra.m: The code for modified Dykstra algorithm (Alg. 2 in the paper)
• sinkhorn_knopp.m: The code for Sinkhorn-Knopp algorithm (Alg. 3 in the paper)
• DCD.m: The code for the DCD method.
• ClusteringMeasure/: Contains calculation of clustering performance metrics (ACC, NMI, PUR, F1)
• Datasets/: Contains the adopted datasets in the paper (except Isolet, COIL100, MNIST).
• NCut9-master/: Contains the implemation of NCut from here).
• loadData.m: Load data from file.
• get_S.m: the code for construction the similarity matrix $S$.