bertinplot: Plot a Bertin Matrix

Description

Plot a data matrix of cases and variables. Each value is represented by a symbol. Large values are highlighted. Note that Bertin arranges the cases horizontally and the variables as rows. The matrix can be rearranged using seriation techniques to make structure in the data visible (see Falguerolles et al 1997).

Usage

# grid-based plot
bertinplot(x, order = NULL, panel.function = panel.bars, highlight = TRUE,
  row_labels = TRUE, col_labels = TRUE, flip_axes = TRUE,
  ...)
# ggplot2-based plot
ggbertinplot(x, order = NULL, geom = "bar", highlight = TRUE,
  row_labels = TRUE, col_labels = TRUE, flip_axes = TRUE,
  prop = FALSE, ...)

Arguments

a data matrix. Note that following Bertin, columns are variables and rows are cases. This behavior can be reversed using reverse = TRUE in options.

order

an object of class ser_permutation to rearrange x before plotting. If NULL, no rearrangement is performed.

panel.function

a function to produce the symbols. Currently available functions are panel.bars (default), panel.circles, panel.rectangles, panel.tiles and panel.lines. For circles and squares neg. values are represented by a dashed border. For blocks all blocks are the same size (can be used with shading=TRUE).

geom

visualization type. Available geometries are: "tile", "rectangle", "circle", "line", "bar", "none".

highlight

a logical scalar indicating whether to use highlighting. If TRUE, all variables with values greater than the variable-wise mean are highlighted. To control highlighting, also a logical matrix or a matrix with colors with the same dimensions as x can be supplied.

row_labels, col_labels

a logical indicating if row and column labels in x should be displayed. If NULL then labels are displayed if the x contains the appropriate dimname and the number of labels is 25 or less. A character vector of the appropriate length with labels can also be supplied.

flip_axes

logical indicating whether to swap cases and variables in the plot. The default (TRUE) is to plot cases as columns and variables as rows.

prop

logical; change the aspect ratio so cells in the image have a equal width and height.

...

ggbertinplot: further parameters are passed on to ggpimage. bertinplot: further parameters can include:

xlab, ylab: labels (default: use labels from x).
spacing: relative space between symbols (default: 0.2).
shading: use gray shades to encode value instead of highlighting (default: FALSE).
shading.function: a function that accepts a single argument in range [.1, .8] and returns a valid corresponding color (e.g., using rgb).
frame: plot a grid to separate symbols (default: FALSE).
mar: margins (see par).
gp_labels: gpar object for labels (see gpar).
gp_panels: gpar object for panels (see gpar).
newpage: a logical indicating whether to start the plot on a new page (see grid.newpage).
pop: a logical indicating whether to pop the created viewports (see pop.viewport)?

Details

The plot is organized as a matrix of symbols. The symbols are drawn by a panel function, where all symbols of a row are drawn by one call of the function (using vectorization). The interface for the panel function is panel.myfunction(value, spacing, hl). value is the vector of values for a row scaled between 0 and 1, spacing contains the relative space between symbols and hl is a logical vector indicating which symbol should be highlighted.

Cut lines can be added to an existing bertin plot using bertin_cut_line(x = NULL, y = NULL). x/y is can be a number indicating where to draw the cut line between two columns/rows. If both x and y is specified then one can select a row/column and the other can select a range to draw a line which does only span a part of the row/column. It is important to call bertinplot() with the option pop = FALSE.

ggbertinplot calls ggpimage and all additional parameters are passed on.

References

de Falguerolles, A., Friedrich, F., Sawitzki, G. (1997): A Tribute to J. Bertin's Graphical Data Analysis. In: Proceedings of the SoftStat '97 (Advances in Statistical Software 6), 11--20.

Examples

Run this code

# NOT RUN {
data("Irish")
scale_by_rank <- function(x) apply(x, 2, rank)
x <- scale_by_rank(Irish[,-6])

# Use the the sum of absolute rank differences
order <- c(
  seriate(dist(x, "minkowski", p = 1)),
  seriate(dist(t(x), "minkowski", p = 1))
)

# Plot
bertinplot(x, order)

# Some alternative displays
bertinplot(x, order, panel = panel.tiles, shading_col = bluered(100), highlight = FALSE)
bertinplot(x, order, panel = panel.circles, spacing = -.2)
bertinplot(x, order, panel = panel.rectangles)
bertinplot(x, order, panel = panel.lines)

# Plot with cut lines (we manually set the order here)
order <- ser_permutation(c(21, 16, 19, 18, 14, 12, 20, 15,
    17, 26, 13, 41,  7, 11, 5, 23, 28, 34, 31, 1, 38, 40,
    3, 39,  4, 27, 24,  8, 37, 36, 25, 30, 33, 35,  2,
    22, 32, 29, 10,  6,  9),
    c(4, 2, 1, 6, 8, 7, 5, 3))

bertinplot(x, order, pop=FALSE)
bertin_cut_line(, 4) ## horizontal line between rows 4 and 5
bertin_cut_line(, 7) ## separate "Right to Life" from the rest
bertin_cut_line(14, c(0, 4)) ## separate a block of large values (vertically)

# ggplot2-based plots
if (require("ggplot2")) {
  library(ggplot2)

  # Default plot uses bars and highlighting values larger than the mean
  ggbertinplot(x, order)

  # highlight values in the 4th quartile
  ggbertinplot(x, order, highlight = quantile(x, probs = .75))

  # Use different geoms. "none" lets the user specify their own geom.
  # Variables set are row, col and x (for the value).

  ggbertinplot(x, order, geom = "tile", prop = TRUE)
  ggbertinplot(x, order, geom = "rectangle")
  ggbertinplot(x, order, geom = "rectangle", prop = TRUE)
  ggbertinplot(x, order, geom = "circle")
  ggbertinplot(x, order, geom = "line")

  # Tiles with diverging color scale
  ggbertinplot(x, order, geom = "tile", prop = TRUE) +
    scale_fill_gradient2(midpoint = mean(x))

  # Custom geom (geom = "none"). Defined variables are row, col, and x for the value
  ggbertinplot(x, order, geom = "none", prop = FALSE) +
    geom_point(aes(x = col, y = row, size = x, color = x > 30), pch = 15) +
    scale_size(range = c(1, 10))

  # Use a ggplot2 theme with theme_set()
  old_theme <- theme_set(theme_minimal() +
      theme(panel.grid = element_blank())
    )
  ggbertinplot(x, order, geom = "bar")
  theme_set(old_theme)
}
# }