forestFloor (version 1.4)

fcol: Generic colour module for forestFloor obejcts

Description

This colour module colour observations by selected variables. PCA decomposes a selection more than three variables. Space can be inflated by random forest variable importance, to focus colouring on influential variables. Outliers(>3std.dev) are automatically supressed. Any colouring can be modified.

Usage

fcol(ff, cols = NULL, orderByImportance = NULL,  X.matrix = TRUE, 
     hue = NULL, saturation = NULL, brightness = NULL,
     hue.range  = NULL, sat.range  = NULL, bri.range  = NULL,
     alpha = NULL, RGB = NULL, max.df=3,
     imp.weight = NULL, imp.exp = 1,outlier.lim = 3,RGB.exp=NULL)

Arguments

ff

a obejct of class "forestFloor" or a matrix or a data.frame. No missing values. No factors(for now).

cols

vector of indices of columns to colour by, will refer to ff$X if X.matrix=T and else ff$FCmatrix. If ff itself is a matrix or data.frame, indices will refer to these coloums

orderByImportance

logical, should cols refer to X column order or columns sorted by variable importance. Input must be of forestFloor -class to use this. Set to FALSE if no importance sorting is wanted. Otherwise leave as is.

X.matrix

logical, true will use feature matrix false will use feature contribution matrix. Only relvant if input is forestFloor object.

hue

value within [0,1], hue=1 will be exactly as hue = 0 colour wheel settings, will skew the colour of all observations without changing the contrast between any two given observations.

saturation

value within [0,1], mean saturation of colours, 0 is greytone and 1 is maximal colourfull.

brightness

value within [0,1], mean brightness of colours, 0 is black and 1 is lightly colours.

hue.range

value within [0,1], ratio of colour wheel, small value is small slice of colour whell those little variation in colours. 1 is any possible colour except for RGB colour system.

sat.range

value within [0,1], for colouring of 2 or more variables, a range of saturation is needed to obtain more degrees of freedom in the colour system. But as saturation of is preferred to be >.75 the range of saturation cannot here exceed .5. If NULL sat.rang

bri.range

value within [0,1], for colouring of 3 or more variables, a range of brightness is needed to obtain more degrees of freedom in the colour system. But as brightness of is preferred to be >.75 the range of saturation cannot here exceed .5. If NULL bri.rang

alpha

value within [0;1] transparency of colours.

RGB

logical TRUE/FALSE, RGB=NULL: will turn TRUE if one variable selected RGB=TRUE: Red-Green-Blue colour: a system with fewer colours(~3) but more contrast. Can still be altered by hue, saturation, brightness etc. RGB=FALSE: True-colour-system: Maximum

max.df

integer 1, 2, or 3 only. Only for true-colour-system, the maximal allowed degrees of freedom in a colour scale. If more variables selected than max.df, PCA decompose to request degrees of freedom. max.df = 1 will give more simple colour gradients

imp.weight

Logical?, Should importance from a forestFloor object be used to weight selected variables? obviously not possible if input ff is a matrix or data.frame. If randomForest(importance=TRUE) during training, variable importance will be used. Otherwise the m

imp.exp

exponent to modify influence of imp.weight. 0 is not influence. -1 is counter influence. 1 is linear influence. .5 is square root influence etc..

outlier.lim

number from 0 to Inf. Any observation which univariately exceed this limit will be suppressed, as if it actually where on this limit. Normal limit is 3 standard deviations. Extreme outliers can otherwise reserve alone a very large part of a given linear c

RGB.exp

value between ]1;>1]. Defines steepness of the gradient of the RGB colour system Close to one green midle area is missing. For values higher than 2, green area is dominating

Value

• a character vector specifying the colour of any observations. Each elements is something like "#F1A24340", where F1 is the hexadecimal of the red colour, then A2 is the green, then 43 is blue and 40 is transparency.

Details

fcol produces colours for any observation. These are used plotting.

Examples

library(forestFloor)
obs=4000 
vars = 6 
X = data.frame(replicate(vars,rnorm(obs))) 
Y = with(X, X1^2 + sin(X2*pi) + 2 * X3 * X4 + 0.5 * rnorm(obs)) 

#grow a forest, remeber to include inbag
rfo=randomForest::randomForest(X,Y,keep.inbag=TRUE,
importance=TRUE,sampsize=700)

#compute topology
ff = forestFloor(rfo,X)

#print forestFloor
print(ff) 

#plot partial functions of most important variables first
Colours1=fcol(ff,1)
plot(ff,plot_seq=NULL,col=Colours1)

#try to colour by first four variables, uses PCA du reduce system to 3-way gradient
# (2.5 way more exactly as saturation and brightness by default have very limited ranges
# to avoid gray - or overexposed color tones).
Colours2=fcol(ff,1:4)
plot(ff,plot_seq=NULL,external.col=Colours2)