drawing a circle with the coefficients shown as an arrow between the variables in ggplot 2

Question

Suppose I have the following data frame:

df <- data.frame(A1 = c(0,3.5,0,2.1), A2  =c(0.9,0,0,0.6), A3 =      c(0,0.3,0,0.3),A4= c(0,1.9,0,0))
rownames(df) <- names(df)

every element df(i,j) is the strength of relation between ith column and jth row (they are mutually connected, meaning strength between 1 to j is different from strength between j to i). A "0" entry means there is no relation.

Now I would like to draw a circle, with the variables on the perimeter of the circle, and an arrow that shows which variables are connected to each other, and hopefully show the strength of the connection based on the width of the arrow.

So, the final product I wish to be something like this:

Is it even possible to do something like that with ggplot2?

Thanks in advance.

Yes, it's possible. Perhaps you can share with us what you have tried already? This will allow us to help you with specific problems you have in drawing your plot. — Weihuang Wong, Dec 08 '16 at 03:11
@WeihuangWong I really have no idea how to proceed. The only helpful information I found on the net is from this link: http://stackoverflow.com/questions/6862742/draw-a-circle-with-ggplot2. It just shows how to draw a circle. Thanks — H_A, Dec 08 '16 at 04:27

score 1 · Answer 1 · answered Dec 08 '16 at 04:32

1

I wrote a little package that does this kind of thing. Here's a small demo vignette https://github.com/mkearney/lavplot/blob/master/vignettes/demo.Rmd. Image of plot provided below.

answered Dec 08 '16 at 04:32

mkearney

1,266
10
18

Thanks Mike. This is useful, but not exactly what I need. It requires a lot of effort when the number of nodes/arrows is large. – H_A Dec 08 '16 at 05:14
You can try the semplot package. Lots more options and useful for larger models. – mkearney Dec 08 '16 at 05:31

Weihuang Wong · Accepted Answer · 2016-12-08T07:53:01.150

igraph

We start by making a graph from your adjacency matrix:

df <- t(df)
ga  <- graph.adjacency(as.matrix(df), weighted = TRUE, mode = "directed")

Then, plot a circle:

par(mar = rep(0.25, 4))
pts <- seq(0, 2*pi, l = 100)
plot(cbind(sin(pts), cos(pts)), type = "l", frame = F, xaxt = "n", yaxt = "n")

Finally, plot the graph:

plot.igraph(ga,
  vertex.label = V(ga)$name,
  edge.width = E(ga)$weight,
  edge.curved = TRUE,
  edge.label = E(ga)$weight,
  layout = layout_in_circle(ga, order = V(ga)), 
  add = T)

Output below. You can customize your graph (e.g. curvature and colors of edges, shapes of vertices) as desired.

ggplot2

The main idea is to set up three sets of geoms: the circle, the nodes (vertices), and the lines (edges). First, we load some packages, and prep the circle and nodes:

library(ggplot2)
library(tidyr)
library(dplyr)

# For circle
pts <- seq(0, 2*pi, l = 100)

# For nodes
theta <- seq(0, 2*pi, l = nrow(df) + 1)[1:nrow(df)]
l <- data.frame(x = sin(theta), y = cos(theta), v = names(df), 
  stringsAsFactors = FALSE)

The edges are a little bit more involved. I make a function to make coordinates for the lines, given an origin and destination:

make_edge <- function(origin, dest, l, shrink = .9) {
  # l is the layout matrix for the nodes that we made previously
  data.frame(
    x0 = l$x[l$v == origin],
    y0 = l$y[l$v == origin],
    x1 = l$x[l$v == dest],
    y1 = l$y[l$v == dest]
  ) * shrink
}

Then, we make an adjacency graph, and bind the edge coordinates to it:

gr <- gather(mutate(df, dest = names(df)), origin, wt, -dest)
gr <- gr[gr$wt != 0, ]
edges <- do.call(rbind, 
  mapply(make_edge, gr$origin, gr$dest, list(l), shrink = .94, SIMPLIFY = F)
  )
ga <- cbind(gr, edges)

Finally, we plot:

ggplot() +
  geom_path(data = data.frame(x = sin(pts), y = cos(pts)), aes(x, y)) +
  geom_label(data = l, aes(x, y, label = v)) +
  geom_curve(data = ga, 
    aes(x = x0, y = y0, xend = x1, yend = y1, size = wt, colour = origin),
    alpha = 0.8,
    curvature = 0.1,
    arrow = arrow(length = unit(2, "mm"))) +
  scale_size_continuous(range=c(.25,2), guide = FALSE) +
  theme_void()

Output:

Pretty neat work specially on how you used ggplot. Thanks a lot. — H_A, Dec 08 '16 at 15:38
Is it possible to show the weights on the arrows as it is shown on the graph generated by igraph? — H_A, Dec 08 '16 at 16:58
Yes, but perhaps it may be better for you to open a separate question. The basic idea is the same as that used to draw the lines. You simply have to augment the `ga` dataframe with the coordinates for the labels (the weights) and use a `geom_text` to plot them. — Weihuang Wong, Dec 08 '16 at 23:59

drawing a circle with the coefficients shown as an arrow between the variables in ggplot 2

2 Answers2

igraph

ggplot2