Parallel coordinates in ggplot2 with ggparcoord

Package

Author

Barret Schloerke

Sample data set

In the examples below we are going to use the iris data set, which consists on different measures about the length and width of the petals and the sepals of flowers of different species.

iris

Most basic parallel coordinate plot with `GGally`

The ggparcoord function from GGally packages allows creating parallel coordinate plots based on ggplot2. To create a basic plot pass the data frame to the function.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris)

Basic paralell coordinate plot in ggplot2

Selecting columns

If you want to select only some columns to be displayed on the X-axis pass a vector of variables to the columns argument.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4)

Parallel coordinates with ggparcoord

Transparency of the lines

The transparency of the lines can be controlled with alphaLines. If there are many lines it is recommended to decrease the level of transparency.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris, 
           columns = 1:4,
           alphaLines = 0.2)

ggparcoord function from GGally

Color by group

If the data set contains a categorical variable representing groups you can pass it to the groupColumn argument to colorize the lines by group.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species")

Parallel coordinate color by group

Color customization

As the ggparcoord is based on ggplot you can change the color of the lines with scale_color_manual or scale_color_brewer, for instance.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species") +
           scale_color_brewer(palette = "Set2")

Parallel lines color customization in R

Adding points

The function provides a logical argument named showPoints which can be set to TRUE to plot the points.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           showPoints = TRUE) +
           scale_color_brewer(palette = "Set2")

Parallel coordinate plot with points in ggplot2

Spline interpolation

The lines can be smoothed via spline interpolation setting splineFactor = TRUE.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           splineFactor = TRUE) +
           scale_color_brewer(palette = "Set2")

Smoothed parallel coordinate lines via spline in R

Adding box plots

Parallel coordinates plot with box plot in ggplot2

It is possible to add box plots for each variable setting boxplot = TRUE.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           alphaLines = 0.2,
           showPoints = TRUE,
           boxplot = TRUE)

Shade box ggparcoord

You can also add a box from the maximum to the minimun for each variable.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           alphaLines = 0.2,
           showPoints = TRUE,
           boxplot = TRUE,
           shadeBox = 4)

Scaling methods

The function provides several scaling methods that can be selected with the scale argument. The default method is "std" and the other possible options are displayed below. Type ?ggparcoord for additional information.

“robust” scale

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           scale = "robust") +
           scale_color_brewer(palette = "Set2")

Robust scale method ggparcoord

“uniminmax” scale

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           scale = "uniminmax") +
           scale_color_brewer(palette = "Set2")

uniminmax scale method in ggparcoord

“globalminmax” scale (no scaling)

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           scale = "globalminmax") +
           scale_color_brewer(palette = "Set2")

No scaling in ggplot2 parallel coordinates plot

“center” scale

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           scale = "center") +
           scale_color_brewer(palette = "Set2")

Center scale ggparcoord

“centerObs” scale

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           scale = "centerObs") +
           scale_color_brewer(palette = "Set2")

centerObs scaling method in ggparcoord

Ordering methods

There are also several methods for ordering the variables displayed on the X-axis. The default ordering is based on the selected columns, but you can also specify a custom order vector or select one of the methods displayed below. Type ?ggparcoord for further details for each mehtod.

Order the variables by their separation between any one class and the rest in ggparcoord

“anyClass” order (by their maximum of k F-statistics)

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           order = "anyClass") +
           scale_color_brewer(palette = "Set2")

coordinates order based on the F-Statistic

“allClass” order (by their F-statistic from an ANOVA)

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           order = "allClass") +
           scale_color_brewer(palette = "Set2")

Order based on the sample skewness

“skewness” order (by their sample skewness)

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           order = "skewness") +
           scale_color_brewer(palette = "Set2")

Parallel coordinate plot in R

“Outlying” order (by the scagnostic measure)

# install.packages("GGally")
# install.packages("scagnostics")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           groupColumn = "Species",
           order = "Outlying") +
           scale_color_brewer(palette = "Set2")

You can make use of the facet_wrap function to create a parallel coordinates plot for each of the groups.

# install.packages("GGally")
library(GGally)

ggparcoord(data = iris,
           columns = 1:4,
           alphaLines = 0.2,
           boxplot = TRUE,
           groupColumn = "Species",
           order = "Outlying") +
           scale_color_brewer(palette = "Set2") +
  facet_wrap(~ Species)

Parallel coordinates by group in ggplot2