OutliersO3 (version 0.6.3)

O3plotM: Draws an Overview of Outliers (O3) plot for more than one method and parallel coordinate plots

Description

Function for drawing Overview of Outliers (O3) plots for comparing outlier methods and for drawing supporting parallel coordinate plots.

Usage

O3plotM(outResults, caseNames=paste0("X", 1:nrow(outResults$data)),
 sortVars=TRUE, coltxtsize=14, O3control=O3plotColours())

Arguments

outResults

a list for each method, and within that for each variable combination, of the variables used, the indices of cases identified as outliers, and the outlier distances for all cases in the dataset.

caseNames

the ID variable used to identify the cases in an O3 plot, the default is the rownames from the dataset (so that they will then just be X7, X11, etc.)

sortVars

sort the variable columns by how often the variables occur in combinations, otherwise keep the variable order in the dataset

coltxtsize

set the size of text for column names in O3 plots (useful if there are so many columns that names overlap)

O3control

A list of colours for O3 plots. If omitted, O3plotColours gives the defaults.

Value

nOut

numbers of outliers found by each method

gpcp

a parallel coordinate plot of the dataset with cases ever found to be outliers coloured red

gO3

an O3 plot

gO3x

an O3 plot for three or more methods in which outliers identified by only one method for a variable combination are ignored.

gMethods

a parallel coordinate plot of the outlier distances calculated by each method for the full dataset with cases ever found to be outliers coloured red

outsTable

a table of all outliers found by case, variable combination, and method. The variable combination labels are a binary coding in the original order of the variables in the dataset.

Cs

a three-dimensional array of methods by variable combinations by cases of the outlier distances calculated.

Details

This function takes the output from O3prep and draws an O3 plot. If there are only two methods, then the default colours are red if both methods identify the case as an outlier for a variable combination and blue or green if only one method does. With more than two methods the default colours are red if all methods identify the outlier, orange if all but one method do, and shades of slategray otherwise.

The two parallel coordinate plots, one using the raw data and one using outlier distances, are examples of what can be done to explore the results in more detail. If you want these plots with other highlighting then you can use outsTable with either the dataset or the Cs array to draw them using ggparcoord from GGally or whatever graphics tool you prefer.

The plots produced are ggplot objects so that you can work with them<U+2014>to some extent<U+2014>directly. In particular, plot margins can be set using + theme(plot.margin = unit(c(t, r, b, l), ''cm'')), which is useful when the cases are labelled with the caseNames option and you need more space to the right of the plot.

See Also

O3plotColours, HDoutliers in HDoutliers, FastPCS in FastPCS, mvBACON in robustX, adjOutlyingness in robustbase, DDC in cellWise, covMcd in robustbase

Examples

# NOT RUN {
c1 <- O3prep(stackloss, k1=2, method=c("HDo", "BAC"), tolHDo=0.025, tolBAC=0.01)
c2 <- O3plotM(c1)
c2$nOut
c2$gpcp
c2$gO3

# }
# NOT RUN {
b1 <- O3prep(stackloss, method=c("HDo", "BAC", "DDC"), tolHDo=0.025, tolBAC=0.01, tolDDC=0.05)
b2 <- O3plotM(b1)
b2$nOut
b2$gpcp
b2$gO3
b2$outsTable
# }
# NOT RUN {
# It is advisable with large datasets to check the number of outliers identified (nOut)
# before drawing graphics.  Occasionally methods find very many outliers.
# }
# NOT RUN {
data(diamonds, package="ggplot2")
data <- diamonds[1:5000, c(1, 5, 6, 8:10)]
pPa <- O3prep(data, method=c("PCS", "adjOut"), tolPCS=0.01, toladj=0.01, boxplotLimits=10)
pPx <- O3plotM(pPa)
pPx$nOut
# }