In tkonopka/MultiPattern: Multi-Pattern Discovery Framework

This workbook generates figures shown on the github landing page (README.md)

To create the figures execute library("rmarkdown"), then render("readme.Rmd").

## Some settings for the vignette
library("knitr")
knitr::opts_chunk$set(cache.path='cache/readme_')
knitr::opts_chunk$set(fig.path='figures/readme_')
knitr::opts_chunk$set(fig.align='center') 

## load packages
library("cluster")
library("MultiPattern")

## Prep for plotting
library("Rcssplot")
MPcss = Rcss("../vignettes/MPvignette.Rcss")
RcssDefaultStyle <- MPcss
RcssOverload()
Rcsspar()
set.seed(12345)

## This block has handy

## split points into cluster usings pam
pamclusters = function(dd, prefix="A",  kmax=3) {
  dpam = pam(dd, k=kmax)
  setNames(paste0(prefix, dpam$clustering), names(dpam$clustering))
}

## perform a customized MultiPattern workflow
mpreadme = function(dd, pamk=2, repr=5) {
  mp = MPnew(rownames(dd), data=list("dd"=dd[,1:2]))
  ## speed up calculation with fewer random configs
  MPchangeSettings(mp, list(num.random=10))
  MPeasyConfig(mp, type=c("pca", "euclidean", "hclust", "pam"))
  metasims = MPgetAverageMetaDistance(mp, verbose=FALSE)
  metamap = MPgetMap(metasims)
  ## get representatives
  if (class(repr)=="integer" | class(repr)=="numeric") {
    repr=round(repr)
    mpreps = MPgetRepresentatives(dist(metamap), method="extreme", k=repr)
    mpreps = grep("rnorm", mpreps, inv=T, val=T)
    mpreps = setNames(mpreps, LETTERS[1:length(mpreps)])
  } else {
    mpreps = setNames(repr, LETTERS[1:length(repr)])
  }
  mpsims = MPgetDistances(mp, configs=mpreps)
  ## get labels
  mpclust = matrix(NA, ncol=length(mpreps), nrow=nrow(dd))
  rownames(mpclust) = rownames(dd)
  colnames(mpclust) = names(mpreps)
  for (i in names(mpreps)) {
    temp = pamclusters(mpsims[[mpreps[i]]], prefix=i, kmax=pamk)
    mpclust[names(temp), i] = temp  
    rm(temp, i)
  }
  ## output all the data
  list(data=dd, mp=mp, metasims=metasims, metamap=metamap, 
       mpreps=mpreps, sims=mpsims,  mpclust=mpclust)  
}


## plot split of a dataset into clusters
## expects xx as a list as output by mpreadme above
plotAC = function(xx, ww=3) {
  par(mfrow=c(1,ww))
  for (i in 1:ww) {
    MPplotScatterWithK(xx$data[,1:2], xx$mpclust[,i], main="", RCC="scatter")
  }
}

Toy datasets

par(mfrow=c(1,2))
MPplotmap(MPdata4S[,c("D1", "D2")], xypadding=0.13)
MPplotmap(MPdata6S[,c("D1", "D2")], xypadding=0.13)

Multi-pattern workflows for toy datasets

(This workflow uses some custom functions defined in this workbook with echo=FALSE)

A dataset with four groups:

repr4 = c("dd:clust.A2reg", "dd:clust.A2alt", "dd:clust.P2alt")
mp4S = mpreadme(MPdata4S, repr=repr4)

A dataset with six groups:

repr6 = c("dd:clust.P3reg", "dd:clust.A3alt", "dd:clust.A2reg", 
      "dd:clust.P3alt","dd:clust.A2alt")
mp6S = mpreadme(MPdata6S, pamk=3, repr=repr6)

Show metamaps

par(mfrow=c(1,2))
MPplotmap(mp4S$metamap, label=T, main="mp4S")
MPplotmap(mp6S$metamap, label=T, main="mp6S")

Visualization

Alternate clusterings for dataset MPdata4S

repr4

plotAC(mp4S, 3) 

Alternate clusterings for dataset MPdata6S

repr6

plotAC(mp6S, 5)

tkonopka/MultiPattern documentation built on May 31, 2019, 3:45 p.m.