FSPmix_Parallel: Implements FSPmix in parallelised manner (splits load across...

In MarcelaCespedes/FSPmix: Exploratory tool for fast unsupervised features selection and clustering using mixture models

Description

Implements FSPmix using parallelisation. Prior to calling FSPmix_Parallel(), user must load required parallelisation packages and allocate the number of cores. FSPmix will be implemented with workload split between the number of allocated CPUs (workers). Input requires the data and parameters required to enable FSPmix algorithm, this function returns a list of the combined output from all workers.

Usage

FSPmix_Parallel(dat, boot.size = NULL, no.bootstrap=NULL)

Arguments

dat	Data frame (required) of the number of features to analyse as columns, number of observations as rows
boot.size	Integer, size of bootstrap sample. If boot.size = NULL, then by default the bootstrap sample becomes floor(0.8*dim(dat)[1])
no.bootstrap	Integer, number of times to perform bootstrapping per feature. If no.bootstrap = NULL, then by default no.bootstrap = 500

Value

	List with length of the number of features containing the following objects
two.groups	Boolean variable denoting if two groups were found, two.groups = TRUE, or otherwise; two.groups = FALSE
summ.op	Data frame summary of the feature, including indicator variable if two groups were found (two.groups), summary statistics of the seperation interval, as well as bootstrapped means of the mixture model components
SampDat_Store	Data frame with bootstrapped samples
Plot	ggplot object which shows the overlapped bootstrapped densities in black, bootstrapped mixture model component means (in gray) as well as the threshold interval (values within the blue lines).
Classification.Pred	Data frame containing the classification of each observation (or person) if two.groups = TRUE. Classification categories are Pred.A and Pred.B according to their feature values determined by FSPmix; and Pred.C denoting an observation remains unclassified as their feature value lies within the seperation interval

Author(s)

Marcela Cespedes

Examples

# Generate data from simulation  study
no.ppl = 500
op<- SimulationStudy1(sEEd = 948575, no.ppl = no.ppl)

#x11()
#op$plot.ColourCoded.genes
#op$plot.WhatFSPmixSees

# set up simulation variables for FSPmix_Sim
feature.dat<- op$dat[, 1:20]
boot.size<- round(no.ppl*0.8) # set bootstrap size sample
no.bootstrap<- 300


library(doParallel)
# Create and registed cluster. See ?makeCluster for further details
# Below we allocate 2 CPU's for

cl<-makeCluster(2, outfile="MulticoreLogging.txt")
cl
registerDoParallel(cl)

# Run FSPmix
sim.op<- FSPmix_Parallel(dat=feature.dat,
                         boot.size = boot.size,
                         no.bootstrap= no.bootstrap)

length(sim.op) # length of number of Features to analyse
class(sim.op)

# Show which features FSPmix found two groups
plot.all<- list()
for(i in 1:20){
  tmp.op<- sim.op[[i]]
  plot.all[[i]]<- tmp.op$Plot
  print(tmp.op$two.groups)
}

multiplot(plotlist = plot.all, cols = 5)

MarcelaCespedes/FSPmix documentation built on May 12, 2020, 5:49 p.m.