Plot many time series in parallel
Description
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.
Usage
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28  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 = brewer.pal(n = 2 * nbands, name = "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 = brewer.pal(n = 2 * nbands, name = "RdYlBu"),
origin)
prepanel.horizonplot(x, y, ..., horizonscale = NA,
nbands = 3L,
origin = function(y) na.omit(y)[1])

Arguments
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. 
Details
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. Iforigin
is specified then that will be constant across panels; otherwise it defaults to the initial value. 
horizonscale
specified butorigin
unspecified: the origin defaults to the initial value in each panel, and colors represent deviations from it in steps ofhorizonscale
(up to 3 steps each way). both
horizonscale
andorigin
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).
Value
An object of class "trellis"
. The
update
method can be used to
update components of the object and the
print
method (usually called by
default) will plot it on an appropriate plotting device.
Warning
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 overrides
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.
Author(s)
Felix Andrews felix@nfrac.org
References
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
See Also
Lattice
,
xyplot.ts
,
panel.xyarea
Examples
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77  ## 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 cutandstack 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"))
