qparallel: Draw a parallel coordinates plot

Description

This function creates a parallel coordinates plot (par-coords) for variables in a data, with each line representing a row.

Usage

qparallel(vars = ~., data, scale = "range", names = break_str(vars),  na.action = na_impute, center = NULL, order = c("none", "MDS", "ANOVA",  "randomForest"), horizontal = FALSE, glyph = c("auto", "line",  "tick", "circle", "square", "triangle"), boxplot = FALSE, width = NULL,  jitter = NULL, amount = NULL, main = "", alpha = 1)

Arguments

scale

data standardizing method; possible values are 'range' (scale columns individually to [0, 1]), 'I' (do not transform; use original values), 'sd' (make each column of mean 0 sd 1), and 'global' (scale all the columns to [0, 1] using global minimum and maximum); other character strings here means to use custom functions (see examples below)

names

the variable labels to use in the plot (by default, they are the variable names with non-alphanumeric characters replaced by line breaks '\n')

na.action

the function to deal with missing values

center

the function to calculate where to center all the variables (e.g. center at the medians), or a numeric value, or NULL (do not center)

order

methods to reorder the variables; see reorder_var

horizontal

logical: direction of axes (horizontal or vertical)

glyph

draw complete segments for all observations or other types of glyphs to represent observations (the latter has speed gain in case of large data)

boxplot

logical: overlay boxplots on top of the par-coords plot or not

width

width of boxplots

jitter

NULL (no jittering) or a character vector to jitter variables (usually those categorical variables)

amount

jitter amount

main

title of plot

alpha

value for alpha-blending

vars

variables to show in the plot: a character vector of variable names, or a numeric vector of column indices, or a two-sided formula like ~ x1 + x2 + x3 (without the left-hand side); see var_names

data

a mutaframe created by qdata

Value

A par-coords plot

Details

See common_key_press for a series of common interactions. Interactions specific to par-coords include: press R to toggle the min/max labels; the arrow keys are used to adjust the order of the variables and flip the values of variables (like a mirror reflection) when the axes are selected by the brush.

Examples

Run this code

library(cranvas)

### (1) flea data
data(flea, package = "tourr")
qflea <- qdata(flea, color = species)

## input variables
qparallel(~., data = qflea)  # all variables
qparallel(~aede1 + aede2 + aede3, data = qflea)  # 3 variables
qparallel(1:4, data = qflea)  # first 4 variables
# input variables by names
qparallel(c("head", "aede1", "tars2", "aede3"), data = qflea)

## scaling
qparallel(~., data = qflea, main = "scale columns individually to [0, 1]")
qparallel(~., data = qflea, scale = "I", main = "unscaled data")
qparallel(~., data = qflea, scale = "sd", main = "mean 0 sd 1")
qparallel(~., data = qflea, scale = "global", main = "scale globally to [0, 1]")

## centering by median; add boxplots to assist understanding
qparallel(~., data = qflea, center = median, boxplot = TRUE)

## ordering
qparallel(~., data = qflea, order = "MDS")  # similar variables together
# color is covariate, order by ANOVA p-value
qparallel(~., data = qflea, order = "ANOVA")
## we can use arrow keys to manually order the variables

## horizontal direction
qparallel(~., data = qflea, horizontal = TRUE)

## use glyphs instead of segments
qparallel(~., data = qflea, glyph = "tick")
qparallel(~., data = qflea, glyph = "circle")

## jittering
qparallel(~., data = qflea, jitter = "species")

## categorical linking
id <- link_cat(qflea, "species")
## now brush on plots; will see all rows in the same category brushed

## stop categorical linking
remove_link(qflea, id)

### (2) NRC rankings
qnrc <- qdata(nrcstat, color = RegCode)

## Overview: type, rankings
qparallel(13:10, data = qnrc, main = "Overview of Rankings", glyph = "tick", 
    horizontal = TRUE, boxplot = TRUE)

## link to a droplist (institution names)
record_selector(Institution, qnrc)

## TODO: we need keyboard interactions here instead of command line
brush(qnrc, "persistent") <- TRUE  # begin persistent brushing
brush(qnrc, "color") <- "brown"
## select other objects now
brush(qnrc, "color") <- "green"
## again, select other objects
brush(qnrc, "color") <- "yellow"
brush(qnrc, "persistent") <- FALSE  # transient brushing

qparallel(vars = 14:19, data = qnrc, main = "Research, Student Support, Diversity", 
    center = median, horizontal = TRUE, glyph = "tick")


### (3) Missing values are imputed by 20% below the mean
df <- as.data.frame(replicate(5, sample(c(rep(NA, 10), rnorm(100)))))
mf <- qdata(df)
qparallel(~., data = mf)

## see this by missing value plot
qmval(~., data = mf)



### (4) alpha transparency
if (require("animation")) {
    data(pollen, package = "animation")
    qpollen <- qdata(pollen)
    qparallel(~., data = qpollen)
    # hold the minus key (-) till the plot is semi-transparent
}



### (5) pressure test
test.mat <- qdata(matrix(rnorm(3e+05 * 10), ncol = 10))
qparallel(~., data = test.mat)
## for large data, short ticks are automatically used instead of segments

cranvas_off()

Run the code above in your browser using DataCamp Workspace