bestPartitions: Best number of cluster partitions.

In nguforche/UnsupRF: Unsupervised Random Forest Clustering

Description

Get "best" number of clustering partitions determined by successiveOmitClusterValidation for k = 2, ..., K using cluster connectivity. Specifically, this function runs successiveOmitClusterValidation for each cluster number in krange and selects the optimal number of cluster by looking at the clustering with the best separation or compactnest

Usage

bestPartitions(data, krange = 2:10, cluster.method = "UnsupRF",
  dist.method = "euclidean", parallel = FALSE, mc.cores = 2,
  control = NULL, sep.measure = "SW", ensemble = TRUE)

Arguments

data	data matrix to cluster
krange	integer vector. Numbers of clusters to try
dist.method	Use RFdist or R's base dist method to compute dissimilarity matrix. Default is "UnsupRF".
parallel	run in parallel ? Currently uses mcapply which works only on linux operating systems
mc.cores	number of CPU cores
control	list with control parameters: ntree number of trees for RFdist no.rep number of repetitions or forests for RFdist neighbSize number of nearest neighbors to compute connectivity validation measure. See the function clusterConnectivity method hclust method RF.parallel type of parallization for RFdist compact.measure one of five compactness measure. See clusterCompactness combined: if ensemble = FALSE, should the seperation measures be combined?
ensemble	take the ensemble of the seperation measures ?

Details

This function is experimental .. work in progress to validate the procedure

Value

clustering for data

nguforche/UnsupRF documentation built on May 5, 2019, 4:51 p.m.