circlepack layout struggles with large number of nodes

[jhjlee]

Hi Thomas,

Thanks for a great tool! I have a question about the circlepack layout that I'm trying to use on my data that has a single level of hierarchy (hundreds of groups, ~15 groups containing 100 or more nodes). When trying to plot, it runs for a while, and eventually throws a "enc3 error" , which I believe is related to the size of the enclosing circle? Is there an upper limit to the number of nodes that can be included in a circlepack plot? Thank you very much.

[schochastics]

Can you post a reproducible example? That would make it easier to check what is going on

[dn-ra]

Hi David,

I'm working with @jhjlee on this data.

Here is a reproducible example that approximates the data we have.

Unfortunately it isn't consistently reproducible. I can run it maybe 5 times, and twice it will execute properly, once it will produce an empty plot, and twice it will result in the problem Hyun Jae describes above. It will hang for a very long time and then produce an enc3 error, or a nonsensical plot. I haven't been able to get it to spit out the enc3 error with this toy data yet, but the hanging will happen after a few tries.

I note that it's normal for it to take some time to actually print out the plot, but creating the plot with ggraph(graph, layout = 'circlepack', weight = size) + geom_node_circle(aes(fill = label, color = factor(depth))) is where the issue comes from.

It seems there is some randomness to the results as it figures out the plot co-ordinates. Does this have something to do with the inefficiencies of the layout algorithms as per #234 ? And are there any solutions you can suggest?

Thanks, Dan

#n of groupings is 50
#n of items is 3204
#n of labels is 10

library(igraph)
library(ggraph)
library(dplyr)
library(magrittr)
library(uuid)
library(RColorBrewer)

items <- uuid::UUIDgenerate(n = 3204) #identifiers of invidual items
labels <- c('boston','brisbane','london','paris','newyork','tokyo','moscow','bogota','beijing','johannesburg') #categories for colouring individual items
groupings <- replicate(paste(sample(LETTERS, 4, TRUE), collapse = ''), n = 50) #grouping variable to pack items in

data <- data.frame(items = items, label = sample(labels, 3204, replace = T), group = sample(groupings, 3204, replace = T)) %>% group_by(group)

#root edges
root_edges =  data.frame('a' = 'root', 'b' = group_keys(data)$group)

#item edges
item_edges = mapply(FUN = function(x, y) {
  cbind('a' = x, 'b' = y$items) } ,
  group_keys(data)$group, group_split(data), USE.NAMES = F
) %>% with(do.call(rbind,.)) %>% as.data.frame()

#this object contains all edges for circlepack plot. items within groupings within 'root'
edge_list = as.matrix(rbind(root_edges, item_edges))

#need to add entries in the original data to accommodate `grouping` and `root` as nodes, so get list of missing node names. This is just for code functionality
missing_node_names = data.frame('items' = do.call(c, root_edges) %>% unique())

#this object contains all the metadata required for the graph
vertex_meta <- data %>% mutate(size = 1) %>% full_join(missing_node_names)  %>% 
  mutate(label = replace(as.character(label), is.na(label), 'A-None'), group = replace(as.character(group), is.na(group), 'A-None')) %>% 
  mutate(size = replace(size, is.na(size), 0),
         label = factor(label, exclude = NULL) )

#init graph
graph = igraph::graph_from_data_frame(edge_list, vertices = vertex_meta)

#colors of lables
label_colors = RColorBrewer::brewer.pal(n = 10, name = 'BrBG')
#colors need to be a named vector for plotting to work
names(label_colors) = levels(as.factor(data$label))
#Add background color of white for all `A-None` data points
colors_for_labels = c("A-None" = 'white', label_colors)

#now create plot
graph_by_label = ggraph(graph, layout = 'circlepack', weight = size) + 
  geom_node_circle(aes(fill = label, color = factor(depth))) +
  theme_void() + coord_fixed() +
  scale_fill_manual(values=colors_for_labels, breaks = names(label_colors)) +
  scale_color_manual(values = c("0" = "white", "1" = "black", "2" ="black"), breaks = c()) + guides(fill = guide_legend('Labels'))

#and show
graph_by_label

[thomasp85]

Thanks - I can reproduce the behaviour based on the provided code. Will investigate

[dn-ra]

Hi Thomas, thanks for fixing. Out of interest could you let me know what caused it?

Thanks, Dan

[thomasp85]

it was due to rounding errors that lead to catastrophic failures as it is so often with computational geometry. It was fixed in the D3 version some time ago but I hadn't kept up to date with it