geom_ribbon() > geom_area() speed difference

[japhir]

I was recently making some figures in which I wanted a simple geom_area() call on something with quite a few points. However, I noticed that my computer was waiting for a long time when calling the function. I had previously plotted something with geom_ribbon() instead, and it was almost instant there. After playing around, it seems that by default on my machine geom_area() is way slower than geom_ribbon() for some reason.

See below for an attempt at a reprex. Benchmark results are a bit iffy though.

library(ggplot2)
library(tibble)

dat <- tibble(
  x = 1:1e4,
  y = rnorm(seq_along(x)) + 5
)

dat |>
  ggplot() +
  geom_area(aes(x = x, y = y))

dat |>
  ggplot() +
  geom_ribbon(aes(x = x, ymin = 0, ymax = y))

# they seem identical
res <- bench::mark(
  area = dat |>
    ggplot() +
    geom_area(aes(x = x, y = y)),
  ribbon = dat |>
    ggplot() +
    geom_ribbon(aes(x = x, ymin = 0, ymax = y)),
  check = FALSE)

# until you actually print them!
res <- bench::mark(
  area = dat |>
    ggplot() +
    geom_area(aes(x = x, y = y)) |> print(), # sloooww
  ribbon = dat |>
    ggplot() +
    geom_ribbon(aes(x = x, ymin = 0, ymax = y)) |> print(), # fast
  check = FALSE)
#> mapping: x = ~x, y = ~y 
#> geom_area: na.rm = FALSE, orientation = NA, outline.type = upper
#> stat_align: na.rm = FALSE, orientation = NA
#> position_stack 
#> mapping: x = ~x, ymin = 0, ymax = ~y 
#> geom_ribbon: na.rm = FALSE, orientation = NA, outline.type = both
#> stat_identity: na.rm = FALSE
#> position_identity 
#> mapping: x = ~x, y = ~y 
#> geom_area: na.rm = FALSE, orientation = NA, outline.type = upper
#> stat_align: na.rm = FALSE, orientation = NA
#> position_stack 
#> mapping: x = ~x, y = ~y 
#> geom_area: na.rm = FALSE, orientation = NA, outline.type = upper
#> stat_align: na.rm = FALSE, orientation = NA
#> position_stack 
#> mapping: x = ~x, y = ~y 
#> geom_area: na.rm = FALSE, orientation = NA, outline.type = upper
#> stat_align: na.rm = FALSE, orientation = NA
#> position_stack 

# < omitted output >

#> geom_ribbon: na.rm = FALSE, orientation = NA, outline.type = both
#> stat_identity: na.rm = FALSE
#> position_identity

# geom_area seems to get many more gc/sec?
res
#> # A tibble: 2 × 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 area         2.03ms   2.13ms      406.    44.3KB     6.19
#> 2 ribbon       2.03ms   2.23ms      375.    22.4KB     6.35

^{Created on 2024-03-20 with reprex v2.1.0}

[japhir]

I guess the reason for this is the default position argument, which defaults to "identify" for ribbon and to stack for area, which may make it slow?

If I switch to:

dat |>
  ggplot() +
  geom_area(aes(x = x, y = y), position = "identity")

I seem to get the same speed.

Perhaps the function could check if there are any group or fill/colour aesthetics and if not, change the position argument to "identity"?

[teunbrand]

I can reproduce the issue, though the benchmarks are measuring how long it takes to print(geom_area()) rather than how long it takes to render the plot :)

Here are more relevant benchmarks; for default options, geom_area() takes 2 seconds whereas geom_ribbon() takes 70 milliseconds.

library(ggplot2)

dat <- data.frame(
  x = 1:1e4,
  y = rnorm(1e4) + 5
)

area   <- ggplot(dat) + geom_area(aes(x, y))
ribbon <- ggplot(dat) + geom_ribbon(aes(x, ymin = 0, ymax = y))

ragg::agg_png(tempfile(fileext = ".png"))

res <- bench::mark(
  area   = print(area),
  ribbon = print(ribbon),
  check  = FALSE,
  min_iterations = 5
)
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
print(res)
#> # A tibble: 2 × 13
#>   expression      min median `itr/sec` mem_alloc `gc/sec` n_itr  n_gc total_time
#>   <bch:expr> <bch:tm> <bch:>     <dbl> <bch:byt>    <dbl> <int> <dbl>   <bch:tm>
#> 1 area          2.02s   2.1s     0.469     129MB    16.7      5   178      10.7s
#> 2 ribbon      72.24ms 72.6ms     9.04     24.2MB     3.62     5     2    552.8ms
#> # ℹ 4 more variables: result <list>, memory <list>, time <list>, gc <list>

Both are about 70 milliseconds when stat and position are identity.

area <- ggplot(dat) + 
  geom_area(aes(x, y), stat = "identity", position = "identity")
ribbon <- ggplot(dat) +
  geom_ribbon(aes(x, ymin = 0, ymax = y), stat = "identity", position = "identity")

res <- bench::mark(
  area   = print(area),
  ribbon = print(ribbon),
  check  = FALSE,
  min_iterations = 5
)
print(res)
#> # A tibble: 2 × 13
#>   expression      min median `itr/sec` mem_alloc `gc/sec` n_itr  n_gc total_time
#>   <bch:expr> <bch:tm> <bch:>     <dbl> <bch:byt>    <dbl> <int> <dbl>   <bch:tm>
#> 1 area         70.6ms 72.1ms      13.8    20.1MB     2.30     6     1      435ms
#> 2 ribbon       70.2ms 71.4ms      14.0    19.5MB     5.61     5     2      356ms
#> # ℹ 4 more variables: result <list>, memory <list>, time <list>, gc <list>

^{Created on 2024-03-21 with reprex v2.1.0}

[japhir]

Nice work dude! Yeah I knew something was wrong in my benchmark, should have realized I could've just forced it to write to file ;-).