CARP.RCC: Robust Convex Clustering

In JhZhang-1999/Rcvxclustr: Robust convex clustering: How does the fusion penalty enhance robustness?

Robust Convex Clustering

Description

CARP_RCC performs convex clustering with the CARP algorithm. The starting value and increment step of lambda can be set.

• n is the number of data observations

• p is the number of features

• nK is the number non-zero weights.

Usage

CARP.RCC(
  X,
  phi,
  method,
  lam.begin,
  lam.step,
  rho,
  tau,
  cl_true,
  randmode,
  max.log = 100
)

Arguments

X	The n-by-p data matrix whose rows are being clustered.
phi	The parameter in the Gaussian kernel weights.
method	The method to be used. Choices are cvx_uni, cvx_GK, Rcvx_uni and Rcvx_GK.
lam.begin	The starting value of lambda.
lam.step	The increment step of lambda. Each iteration will mutiply lambda by this step value.
rho	Augmented Lagrangian penalty parameter.
tau	The robustification parameter in huber loss.
cl_true	The true clustering results. Used for rand index calculation.
randmode	The rand index mode. See adjustedRand in package clues for details.
max.log	The maximal number of iterations. The algorithm also stops when the present iteration gives out the result where all data points are classified in the same cluster.

Value

method The method used.

rand The best rand index obtained.

lam The best lambda value, which reaches the best rand index.

path A matrix, of which each row represents the clustering result for each iteration.

cl_est The result of cluster estimation which produces the best rand index.

JhZhang-1999/Rcvxclustr documentation built on Aug. 18, 2022, 9:25 a.m.