ggplot2 vs base-r plotting and its verbose nature

Question

In the question found here OP asks a rather simple question, which according to my belief has a simple answer. However, OP rejects this as being too verbose, and this caught my interest in the question whether ggplot2 is unnecessarily more complex than base R.

Lets say I want to produce the following plot,

I can do this is base R with the following script,

# Grid of X-axis values
x <- seq(0, 10, 0.01)

# Data
set.seed(1)
y1 <- 2 * cos(x) + 8
y2 <- 3 * sin(x) + 4

# Lines
plot(x, y1, type = "l",
     ylim = c(1, 10), ylab = "y")
lines(x, y2, type = "l", col = 2)

# Fill area between lines
polygon(c(x, rev(x)), c(y2, rev(y1)),
        col = "#6BD7AF")

Which equivalently can be expressed by ggplot2-syntax with the following script,

library(tidyverse)

x <- seq(0, 10, 0.01)

# Data
set.seed(1)
y1 <- 2 * cos(x) + 8
y2 <- 3 * sin(x) + 4



sample_data <- tibble(
        x,
        y1,
        y2
)

sample_data %>% 
        ggplot() + 
        geom_ribbon(
                aes(
                x =x, ymin = y1, ymax = y2
                ),
                fill = "#6BD7AF",
                color = "black"
        ) + theme_classic()

In this discussion about the %>%-operator there is mention of some overhead costs, with an suitable example. So I compared the two methods with microbenchmark, to see whether this also was the case here. The results are found below,

Unit: milliseconds
   expr       min        lq      mean    median        uq       max neval
 ggplot  2.836232  3.013435  3.350137  3.188025  3.388644  9.357495   100
  baseR 20.090868 58.090015 58.993941 59.568097 60.235889 65.577104   100

On average ggplot outruns base R by far. The script below produces these results,

library(microbenchmark)


base_foo <- function() {
        
        # Grid of X-axis values
        x <- seq(0, 10, 0.01)
        
        # Data
        set.seed(1)
        y1 <- 2 * cos(x) + 8
        y2 <- 3 * sin(x) + 4
        
        # Lines
        plot(x, y1, type = "l",
             ylim = c(1, 10), ylab = "y")
        lines(x, y2, type = "l", col = 2)
        
        # Fill area between lines
        polygon(c(x, rev(x)), c(y2, rev(y1)),
                col = "#6BD7AF")
        
}


ggplot_foo <- function() {
        
        # Grid of X-axis values
        x <- seq(0, 10, 0.01)
        
        # Data
        set.seed(1)
        y1 <- 2 * cos(x) + 8
        y2 <- 3 * sin(x) + 4
        
        
        sample_data <- tibble(
                x,
                y1,
                y2
        )
        
        sample_data %>% 
                ggplot() + 
                
                geom_ribbon(
                        
                        aes(
                                x =x, ymin = y1, ymax = y2
                        ),
                        fill = "#6BD7AF",
                        color = "black"
                ) + theme_classic()
        
        
        
}


microbenchmark(
        ggplot = ggplot_foo(),
        baseR  = base_foo()
)

ggplot2 has the advantage of being faster in this specific application, but is there cases where it would make more sense to use base R?

Namely,

Is one or the other more memory-efficient than the other?
Is one or the other more compliant to good coding practise, ie. does it have easier readability across languages?

In summary, ggplot is faster. But when would base R be preferable to ggplot when dealing with, say, plots for presentations and policy recommendations.

ggplot2 vs base-r plotting and its verbose nature

0 Answers