forestFloor (version 1.11.1)

plot_simplex3: 3-class simplex forestFloor plot

Description

3-class forestFloor plotted in a 2D simplex. The plot describes with feature contributions the change of predicted class probability for each sample due a single variable given all other variables. This plot is better than regular multiclass plots (plot.forestFloor_multiClass) to show the change of class probabilities, but the feature values can only be depcited as a colour gradient. But (fig3d=TRUE) allows the feature value to be depicted by the Z-axis as a extra pop-up 3D plot.

Usage

plot_simplex3(ff,
              Xi           = NULL,
              includeTotal = TRUE,
              label.col    = NULL,
              fig.cols     = 3,
              fig.rows     = NULL,
              auto.alpha   = 0.25,
              fig3d        = FALSE,
              restore_par  = TRUE,
              set_pars     = TRUE,
              zoom.fit     = NULL,
              var.col      = NULL,
              plot.sep.centroid = TRUE)

Arguments

ff

x also abbrivated ff, forestFloor_mulitClass the output from the forestFloor function. Must have 3 classes exactly.

Xi

vector of integer indices (refeering to column order of trainingset) to what feature contributions should be plotted in individual plots.

includeTotal

TRUE / FALSE. Combined separation of all feature contributions, which is equal to the separation of the entire model can be included.

label.col

a colour vector of K classes length defining the colour of each class for plotting. NULL is auto.

fig.cols

How many columns should be plotted sideways, is passed to par(mfrow=c(fig.rows,fig.cols))

fig.rows

How many rows should be plotted, is passed to par(mfrow=c(fig.rows,fig.cols)) NULL is auto

auto.alpha

a scalar between 0.5 to 1 most often. Low values increase transparancy of points used to avoid overplotting. auto.alpha is alpha corrected of samplesize such that less adjustment is needed.

fig3d

TRUE/FALSE, a 3D plot including the variable as an axis can be co-plotted with rgl.

restore_par

TRUE/FALSE, calls to graphics par() will be reset

set_pars

TRUE/FALSE, if FALSE plot function will rather inherrit plot settings global pars. USeful for multi plotting loops.

zoom.fit

NULL/TRUE, if TRUE zooming on samples will be applied. Do not set to FALSE.

var.col

a single colour or a colour vector of N samples length. Samples will be coloured accordingly. use function fcol to make colour gradient e.g. by the variable values themselves. See example fcol.

plot.sep.centroid

TRUE/FALSE. Should the average bootstrap prediction be plotted? If no bootstrap stratification, the average bootstrap prediction is equal to class distribution training set. RF model probalistic predictions is equal to average bootstrap prediction plus all feature contributions.

Details

Random forest 3 class maps from a feature space to a 3 dimensional (K-1) probability simplex space, which can be plotted in 2D because class probabilities sum to one, and class feature contributions sum to zero. The centroid these plots is the prior of the random forest model. The prior, unless modified with statification is the target class distribution. Default majority voting lines would run from middle to the corners.

Examples

  library(randomForest)
  library(forestFloor)
  require(utils)
  
  data(iris)

  X = iris[,!names(iris) %in% "Species"]
  Y = iris[,"Species"]
  as.numeric(Y)
  rf.test42 = randomForest(X,Y,keep.forest=TRUE,
    replace=FALSE,keep.inbag=TRUE,samp=15,ntree=100)
  ff.test42 = forestFloor(rf.test42,X,calc_np=FALSE,binary_reg=FALSE)
  
  plot(ff.test42,plot_GOF=TRUE,cex=.7,
       colLists=list(c("#FF0000A5"),
                     c("#00FF0050"),
                     c("#0000FF35")))
  
  show3d(ff.test42,1:2,3:4,plot_GOF=TRUE)
  
  #plot all effect 2D only
  pars = plot_simplex3(ff.test42,Xi=c(1:3),restore_par=FALSE,zoom.fit=NULL,
    var.col=NULL,fig.cols=2,fig.rows=1,fig3d=FALSE,includeTotal=TRUE,auto.alpha=.4
    ,set_pars=TRUE)
  
  pars = plot_simplex3(ff.test42,Xi=0,restore_par=FALSE,zoom.fit=NULL,
    var.col=alist(alpha=.3,cols=1:4),fig3d=FALSE,includeTotal=TRUE,
    auto.alpha=.8,set_pars=FALSE)
  
  for (I in ff.test42$imp_ind[1:4])  {
    #plotting partial OOB-CV separation(including interactions effects)
    #coloured by true class
    pars = plot_simplex3(ff.test42,Xi=I,restore_par=FALSE,zoom.fit=NULL,
    var.col=NULL,fig.cols=4,fig.rows=2,fig3d=TRUE,includeTotal=FALSE,label.col=1:3,
    auto.alpha=.3,set_pars = (I==ff.test42$imp_ind[1]))
    
    #coloured by varaible value
    pars = plot_simplex3(ff.test42,Xi=I,restore_par=FALSE,zoom.fit=TRUE,
    var.col=alist(order=FALSE,alpha=.8),fig3d=FALSE,includeTotal=(I==4),
    auto.alpha=.3,set_pars=FALSE)
  }