An Efficient Semismooth Newton based Algorithm for Convex Clustering

Authors: Yancheng Yuan, Defeng Sun, Kim-Chuan Toh

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive numerical experiments on both simulated and real data demonstrate that our algorithm is highly efficient and robust for solving large-scale problems.
Researcher Affiliation Academia 1Department of Mathematics, National University of Singapore 2Department of Applied Mathematics, Hong Kong Polytechnic University 3Department of Mathematics, National University of Singapore.
Pseudocode Yes Algorithm 1 SSNAL for (P), Algorithm 2 SSNCG for (9), Algorithm 3 IADMM for (P)
Open Source Code No The paper mentions using open source software CVXCLUSTR, but does not provide concrete access to their own implementation code (written in MATLAB).
Open Datasets Yes MNIST, Fisher Iris, WINE, Yale Face B(10Train subset)., Fisher. Fisher iris dataset, 1936. UCI Machine Learning Repository https://archive.ics.uci.edu/ml/datasets/iris.
Dataset Splits No The paper discusses solving the convex clustering model for a range of gamma values to generate a clustering path and evaluates performance, but does not specify train/validation/test splits or cross-validation methodology for data partitioning.
Hardware Specification Yes All our computational results are obtained from a desktop having 16 cores with 32 Intel Xeon E5-2650 processors at 2.6 GHz and 64 GB memory.
Software Dependencies No The paper states, 'We write our code in MATLAB without any dedicated C functions.' and mentions using 'CVXCLUSTR' which is 'an R package', but it does not provide specific version numbers for MATLAB, R, or any other software dependencies.
Experiment Setup Yes In the experiments, we choose k = 10, φ = 0.5 (for the weights wij) and γ [0.2 : 0.2 : 10] to generate the clustering path. In our experiments, we set ϵ = 10 6 unless specified otherwise. When we generate the clustering path for the first parameter value of γ, we first run the IADMM introduced in Algorithm 3 for 100 iterations to generate an initial point, then we use SSNAL to solve (2).