Accelerating ggplot2: use a canvas to speed up rendering plots

July 3, 2017
By

(This article was first published on Ilya Kashnitsky, and kindly contributed to R-bloggers)

One of the nice features of the ggapproach to plotting is that one can save plots as R objects at any step and use later to render and/or modify. I used that feature extensively while creating maps with ggplot2 (see my previous posts: one, two, three, four, five). It is just convenient to first create a canvas with all the theme parameters appropriate for a map, and then overlay the map layer. At some point I decided to check if that workflow was computationally efficient or not. To my surprise, the usage of canvas reduces the rendering time of a ggplot quite a lot. To my further surprise, this finding holds for simple plots as well as maps.

Let’s start with a simple check.

# load required packages
library(tidyverse)      # data manipulation and viz
library(ggthemes)       # themes for ggplot2
library(viridis)        # the best color palette
library(rgdal)          # deal with shapefiles
library(microbenchmark) # measure the speed of executing
library(extrafont)      # nice font
myfont <- "Roboto Condensed"
library(RColorBrewer)

# create a canvas 
canv_mt <- ggplot(mtcars, aes(hp, mpg, color = cyl))+
        coord_cartesian()

# test speed with mocrobenchmark
test <- microbenchmark(
        without_canvas = ggplot(mtcars, aes(hp, mpg, color = cyl))+
                coord_cartesian()+
                geom_point()
        
        ,
        
        with_canvas = canv_mt+
                geom_point()
       
        ,
        
        times = 100
)

test

autoplot(test)+
        aes(fill = expr)+
        scale_fill_viridis(discrete = T)+
        theme_bw(base_size = 15, base_family = myfont)+
        theme(legend.position = "none",
              axis.text = element_text(size = 15))+
        labs(title = "The speed of rendering a simple plot produced with ggplot2")

fig1
Figure 1. Microbenchmark output for a simple plot

The median time of execution is 2.7 milliseconds for the plot without canvas and 1.9 milliseconds for the plot with canvas.

Next, let’s do the same check for a map. For that, I will use the data prepared for one of my earlier posts and recreate the simple map that shows the division of European Union 27 into three subregions.

fig2
Figure 2. The map we use to test the rendering speed

# load the already prepared data
load(url("https://ikashnitsky.github.io/doc/misc/map-subplots/df-27-261-urb-rur.RData"))
load(url("https://ikashnitsky.github.io/doc/misc/map-subplots/spatial-27-261.RData"))

# fortify spatial objects
neib <- fortify(Sneighbors)
bord <- fortify(Sborders)
fort <- fortify(Sn2, region = "id")

# join spatial and statistical data
fort_map <- left_join(df, fort, "id")

# pal for the subregions
brbg3 <- brewer.pal(11,"BrBG")[c(8,2,11)]

# create a blank map
basemap <- ggplot()+
        geom_polygon(data = neib,
                     aes(x = long, y = lat, group = group),
                     fill = "grey90",color = "grey90")+
        coord_equal(ylim = c(1350000,5450000), 
                    xlim = c(2500000, 6600000), 
                    expand = c(0,0))+
        theme_map(base_family = myfont)+
        theme(panel.border = element_rect(color = "black",size = .5,fill = NA),
              legend.position = c(1, 1),
              legend.justification = c(1, 1),
              legend.background = element_rect(colour = NA, fill = NA),
              legend.title = element_text(size = 15),
              legend.text = element_text(size = 15))+
        labs(x = NULL, y = NULL)


# test speed with mocrobenchmark
test_map <- microbenchmark(
        without_canvas = 
                ggplot()+
                geom_polygon(data = neib,
                             aes(x = long, y = lat, group = group),
                             fill = "grey90",color = "grey90")+
                coord_equal(ylim = c(1350000,5450000), 
                            xlim = c(2500000, 6600000), 
                            expand = c(0,0))+
                theme_map(base_family = myfont)+
                theme(panel.border = element_rect(color = "black",
                                                  size = .5,fill = NA),
                      legend.position = c(1, 1),
                      legend.justification = c(1, 1),
                      legend.background = element_rect(colour = NA, fill = NA),
                      legend.title = element_text(size = 15),
                      legend.text = element_text(size = 15))+
                labs(x = NULL, y = NULL) +
                geom_polygon(data = fort_map, 
                             aes(x = long, y = lat, group = group, 
                                 fill = subregion), color = NA)+
                scale_fill_manual(values = rev(brbg3)) +
                theme(legend.position = "none")
        
        ,
        
        with_canvas = 
                basemap +
                geom_polygon(data = fort_map, 
                             aes(x = long, y = lat, group = group, 
                                 fill = subregion), color = NA)+
                scale_fill_manual(values = rev(brbg3)) +
                theme(legend.position = "none")
        
        ,
        
        times = 100
)
      

autoplot(test_map)+
        aes(fill = expr)+
        scale_fill_viridis(discrete = T)+
        theme_bw(base_size = 15, base_family = myfont)+
        theme(legend.position = "none",
              axis.text = element_text(size = 15))+
        labs(title = "The speed of rendering a map produced with ggplot2")

fig3
Figure 3. Microbenchmark output for a map

The median time of execution is 16 milliseconds for the map without canvas and 5.2 milliseconds for the map with canvas.

Conclusion

Use canvas with ggplot2

For the full script to reproduce the results check out this gist.

To leave a comment for the author, please follow the link and comment on their blog: Ilya Kashnitsky.

R-bloggers.com offers daily e-mail updates about R news and tutorials on topics such as: Data science, Big Data, R jobs, visualization (ggplot2, Boxplots, maps, animation), programming (RStudio, Sweave, LaTeX, SQL, Eclipse, git, hadoop, Web Scraping) statistics (regression, PCA, time series, trading) and more...



If you got this far, why not subscribe for updates from the site? Choose your flavor: e-mail, twitter, RSS, or facebook...

Comments are closed.

Search R-bloggers

Sponsors

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)