Plot many time series in parallel
Plot many time series in parallel by cutting the y range into segments and overplotting them with color representing the magnitude and direction of deviation.
horizonplot(x, data, ...)
## Default S3 method:
horizonplot(x, data = NULL, ...,
nbands = 3L,
horizonscale = NA,
origin = function(y) na.omit(y)[1],
colorkey = FALSE, legend = NULL,
panel = panel.horizonplot,
prepanel = prepanel.horizonplot,
col.regions = hcl.colors(2 * nbands, palette="RdYlBu"),
strip = FALSE, strip.left = TRUE,
par.strip.text = list(cex = 0.6),
colorkey.digits = 3,
layout = c(1, NA),
groups = NULL,
default.scales =
list(y = list(relation = "free", axs = "i",
draw = FALSE, tick.number = 2)))
panel.horizonplot(x, y, ..., border = NA,
nbands = 3L,
col.regions = hcl.colors(2 * nbands, palette="RdYlBu"),
origin)
prepanel.horizonplot(x, y, ..., horizonscale = NA,
nbands = 3L,
origin = function(y) na.omit(y)[1])x, y |
Argument on which argument dispatch is carried out. Typically this will be a multivariate time series. In the panel and prepanel functions, these are the data coordinates. |
data |
Not used (at least, not used by |
... |
further arguments. Arguments to |
nbands |
Integer giving the number of discrete color bands used (for both negative and positive deviations from the origin). |
horizonscale |
the scale of each color segment. There are 3 positive
segments and 3 negative segments. If this is a given as a
number then all panels will have comparable distances, though not
necessarily the same actual values (similar in concept to
|
origin |
the baseline y value for the first (positive) segment
(i.e. the value at which red changes to blue). This can be a
number, which is then fixed across all panels, or it can be a
function, which is evaluated with the |
colorkey, legend |
if |
panel |
function to render the graphic given the data. This is the function that actually implements the display. |
prepanel |
function determining range of the data rectangle from data to be used in a panel. |
col.regions |
color scale, with at least 6 colors. This should be a divergent color scale (typically with white as the central color). |
strip, strip.left |
by default strips are only drawn on the left, to save space. |
par.strip.text |
graphical parameters for the strip text; see
|
colorkey.digits |
digits for rounding values in colorkey labels. |
layout |
Numeric vector of length 2 (or 3) specifying number of columns and rows (and pages) in the plot. The default is to have one column and as many rows as there are panels. |
default.scales |
sets default values of |
groups |
not applicable to this type of plot. |
border |
border color for the filled polygons, defaults to no border. |
This function draws time series as filled areas, with modifications to effectively visualise many time series in parallel. Data that would be drawn off the top of each panel is redrawn from the bottom of the panel in a darker color. Values below the origin are inverted and drawn in the opposite color. There are up to three shades (typically in blue) for data above the baseline and up to three shades (typically in red) for data below the baseline. See the article referenced below for an introduction to Horizon plots.
There are three different cases of using this function:
horizonscale unspecified (default case): then each
panel will have different scales, and the colors represent
deviations from the origin up to the maximum deviation from the
origin in that panel. If origin is specified then that will
be constant across panels; otherwise it defaults to the initial
value.
horizonscale specified but origin unspecified:
the origin defaults to the initial value in each panel, and colors
represent deviations from it in steps of horizonscale (up to
3 steps each way).
both horizonscale and origin specified: each
panel will have the same scales, and colors represent fixed ranges
of values.
In each of these cases the colorkey is labelled slightly
differently (see examples).
Note that the y scale in each panel defines the actual origin and
scale used. The origin and horizonscale arguments are
only used in the prepanel function to choose an appropriate y
scale. The ylim argument therefore over-rides
origin and horizonscale. This also implies that choices
of scales$y$relation other than "free" may have
unexpected effects, particularly "sliced", as these change the
y limits from those requested by the prepanel function.
Felix Andrews felix@nfrac.org
Stephen Few (2008). Time on the Horizon. Visual Business Intelligence Newsletter, June/July 2008 http://www.perceptualedge.com/articles/visual_business_intelligence/time_on_the_horizon.pdf
## generate a random time series object with 12 columns
set.seed(1)
dat <- ts(matrix(cumsum(rnorm(200 * 12)), ncol = 12))
colnames(dat) <- paste("series", LETTERS[1:12])
## show simple line plot first, for reference.
xyplot(dat, scales = list(y = "same"))
## these layers show scale and origin in each panel...
infolayers <-
layer(panel.scaleArrow(x = 0.99, digits = 1, col = "grey",
srt = 90, cex = 0.7)) +
layer(lim <- current.panel.limits(),
panel.text(lim$x[1], lim$y[1], round(lim$y[1],1), font = 2,
cex = 0.7, adj = c(-0.5,-0.5), col = "#9FC8DC"))
## Case 1: each panel has a different origin and scale:
## ('origin' default is the first data value in each series).
horizonplot(dat, layout = c(1,12), colorkey = TRUE) +
infolayers
## Case 2: fixed scale but different origin (baseline):
## (similar in concept to scales = "sliced")
horizonplot(dat, layout = c(1,12), horizonscale = 10, colorkey = TRUE) +
infolayers
## Case 3: fixed scale and constant origin (all same scales):
horizonplot(dat, layout = c(1,12), origin = 0, horizonscale = 10, colorkey = TRUE) +
infolayers
## same effect using ylim (but colorkey does not know limits):
horizonplot(dat, layout = c(1,12), ylim = c(0, 10), colorkey = TRUE) +
infolayers
## same scales with full coverage of color scale:
horizonplot(dat, layout = c(1,12), origin = 0,
scales = list(y = list(relation = "same")),
colorkey = TRUE, colorkey.digits = 1) +
infolayers
## use ylab rather than strip.left, for readability.
## also shade any times with missing data values.
horizonplot(dat, horizonscale = 10, colorkey = TRUE,
layout = c(1,12), strip.left = FALSE,
ylab = list(rev(colnames(dat)), rot = 0, cex = 0.7)) +
layer_(panel.fill(col = "gray90"), panel.xblocks(..., col = "white"))
## illustration of the cut points used in the following plot
xyplot(EuStockMarkets, scales = list(y = "same"),
panel = function(x, y, ...) {
col <-
c("#B41414","#E03231","#F7A99C","#9FC8DC","#468CC8","#0165B3")
for (i in c(-3:-1, 2:0)) {
if (i >= 0)
yi <- pmax(4000, pmin(y, 4000 + 1000 * (i+1)))
if (i < 0)
yi <- pmin(4000, pmax(y, 4000 + 1000 * i))
panel.xyarea(x, yi, origin = 4000,
col = col[i+4], border = NA)
}
panel.lines(x, y)
panel.abline(h = 4000, lty = 2)
})
## compare with previous plot
horizonplot(EuStockMarkets, colorkey = TRUE,
origin = 4000, horizonscale = 1000) +
infolayers
## a cut-and-stack plot; use constant y scales!
horizonplot(sunspots, cut = list(n = 23, overlap = 0),
scales = list(draw = FALSE, y = list(relation = "same")),
origin = 100, colorkey = TRUE,
strip.left = FALSE, layout = c(1,23)) +
layer(grid::grid.text(round(x[1]), x = 0, just = "left"))Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.