KNOBSynC: Kernel-estimated Nonparametric Overlap-Based Syncytial...

In ialmodovar/SynClustR: Syncytial Clustering Algorithms in R

Kernel-estimated Nonparametric Overlap-Based Syncytial Clustering

Description

Merge clustering components using smooth estimation of the missclassification probabilities

Usage

KNOBSynC(x, kmns.results=NULL, Kmax = NULL,EstK=NULL,
kernel = "RIG", kappa = NULL, b = NULL,desired.ncores=2,
trueids=NULL,inv.roles=FALSE,
ret.steps = FALSE,verbose=FALSE,...)

Arguments

x	vector or matrix
kmns.results	Default is NULL. If k-means results is provided will use it, if not will run k-means until Kmax and estimate the number of groups.
Kmax	Default is NULL.If not provided will choose Kmax = max50,sqrt(n)
EstK	If need to estimate the number of groups will use the Jump method of Sugar and James (2003). If EstK = "KL" will estimate the number of groups using Krznaswoki and Lai (1985).
kernel	Choice of the kernel to be use in the smooth estimation. Default is Reciprocal Inverse Gaussian (RIG), other choices are Gamma kernel, Gamma kernel with inverse roles and Gaussian kernel.
kappa	Merging parameter. This defined as kappa * gen.overlap. See Almodovar-Rivera and Maitra (2018).
b	Smoothing parameter (bandwith) to be use in the smooth estimation of the distribution function. If not value is provided will use a estimate of the MISE
desired.ncores	Set the desired number of processor cores to be use. Default is min(desired.ncores, maximum number of cores).
trueids	If provided will compute the adjusted Rand Index at each step
inv.roles	Inverse role when using a gamma kernel estimator, Jeon and Kim (2013). Default is FALSE using Chen (2000).
ret.steps	Return steps calculation for each iteration. Default is FALSE
verbose	FALSE

Show progress in the console.

...	Arguments to be use if k-means.all() need to be call.

Author(s)

Israel Almodovar-Rivera and Ranjan Maitra.

References

Almodovar-Rivera, I., & Maitra, R. (2018). Kernel-estimated Nonparametric Overlap-Based Syncytial Clustering. arXiv preprint arXiv:1805.09505.

Examples

set.seed(787)
## Example 1: Merging will occur. Best clustering solution when generalized overlap = 
data(Bullseye)
oo <- KNOBSynC(x = Bullseye[,-3],verbose = TRUE)
Bullseye$IdsKmeans <- oo$Ids
Bullseye$IdsKNOBSynC <- oo$IdsMerge

par(mfrow=c(1,3))
with(Bullseye,plot(x = x,y = y, col=Ids,main="True"))
with(Bullseye,plot(x = x,y = y, col=IdsKmeans,main="k-means"))
with(Bullseye,plot(x = x,y = y, col=IdsKNOBSynC,main="KNOB-SynC"))

## Example 2 Merging will not occur since generalized-overlap approx max-overlap 
data(Spherical7)
oo <- KNOBSynC(x = Spherical7[,-3],verbose = TRUE)
Spherical7$IdsKmeans <- oo$Ids
Spherical7$IdsKNOBSynC <- oo$IdsMerge

par(mfrow=c(1,3))
with(Spherical7,plot(x = x,y = y, col=Ids,main="True"))
with(Spherical7,plot(x = x,y = y, col=IdsKmeans,main="k-means"))
with(Spherical7,plot(x = x,y = y, col=IdsKNOBSynC,main="KNOB-SynC"))

## Example 3 a more difficult data  take some time to finish

data(SSS)
oo <- KNOBSynC(x = SSS[,-3],verbose = TRUE,nstart = 100)
SSS$IdsKmeans <- oo$Ids
SSS$IdsKNOBSynC <- oo$IdsMerge

par(mfrow=c(1,3))
with(SSS,plot(x = x,y = y, col=Ids,main="True"))
with(SSS,plot(x = x,y = y, col=IdsKmeans,main="k-means"))
with(SSS,plot(x = x,y = y, col=IdsKNOBSynC,main="KNOB-SynC"))

ialmodovar/SynClustR documentation built on July 7, 2023, 12:18 a.m.