odvis is an R package that provides helper functions for visualizing
directed origin-destination data using ggplot2.
You can install the development version of odvis from
GitHub with:
# install.packages("devtools")
devtools::install_github("rogerbeecham/odvis")Standard flow visualizations draw lines between origin-destination
pairs. In ggplot2 this can be achieved by passing coordinates to
geom_path(), one row for origin and destination respectively. In
order to represent both outward and inward directions, we have to offset
the locations (y-position) of the coordinate pairs.
od_pair <- tribble(
~direction, ~o_d, ~x, ~y,
"out", "origin", 0, .2,
"out", "destination", 1, .2,
"in", "origin", 1, -.2,
"in", "destination", 0, -.2,
)
od_pair |>
ggplot() +
geom_point(aes(x=x, y=y*.2, group=direction, colour=direction)) +
geom_path(aes(x=x, y=y*.2, group=direction, colour=direction)) +
geom_text(aes(x=x, y=y*.5, label=o_d, colour=direction)) +
geom_text(data=. %>% filter(o_d=="origin"), aes(x=.5, y=y, label=direction, colour=direction)) +
scale_colour_manual(values=c("#b2182b", "#2166ac"), guide="none") +
coord_equal(ylim=c(-.2, +.2), xlim=c(-.1, +1.1)) The ggforce extension provides
a function for generating various curves, and we can use geom_bezier()
to add a control point to offset curves depending on whether the
direction is outward (bends upward) or inward (bends downward). This
allows us to represent the origin and destination point locations in
their exact position.
od_pair <- tribble(
~direction, ~o_d, ~x, ~y,
"out", "o", 0, .015,
"out", "c", 0.5, .2,
"out", "d", 1,.015,
"in", "o", 1, -.015,
"in", "c", 0.5, -.2,
"in", "d", 0, -.015,
)
od_pair |>
ggplot() +
geom_point(data=. %>% filter(o_d!="c"), aes(x=x, y=0, group=direction)) +
geom_bezier0(aes(x=x, y=y, group=direction, colour=direction)) +
geom_text(data=. %>% filter(o_d!="c"),
aes(x=x, y=y*3, label=o_d, colour=direction)) +
geom_text(data=. %>% filter(o_d=="o"), aes(x=.5, y=y*12, label=direction, colour=direction)) +
scale_colour_manual(values=c("#b2182b", "#2166ac"), guide="none") +
coord_equal(ylim=c(-.2, +.2), xlim=c(-.1, +1.1))A problem here is that the control point is not generalisable and the
symbolisation not intuitive. The outward and inward directions are
differentiated, but imagine a complex graph dataset with many edges in
2D space. So we move the control points such that lines curve towards
the destination. This is achieved with get_trajectory(), which takes a
character string naming the od-pair and locations in x and y of that
OD’s origin an destination. The function returns a tibble of three
rows representing origin, control point and destination respectively.
dat <- bind_rows(
outward=get_trajectory(o_x=0,o_y=0,d_x=1,d_y=0, "out") |>
add_column(type=c("o","c","d")),
inward=get_trajectory(o_x=1,o_y=0,d_x=0,d_y=0, "in") |>
add_column(type=c("o","c","d"))
)
dat |>
ggplot(aes(x=x,y=y, colour=od_pair)) +
geom_point(data=. %>% filter(type!="c"), colour="#252525") +
ggforce::geom_bezier0(aes(group=od_pair)) +
geom_text(data=. %>% filter(type=="c"), aes(label=od_pair)) +
scale_colour_manual(values=c("#b2182b", "#2166ac"), guide="none") +
coord_equal() 