Lecture 25: Making code more efficient

Approaches to faster code

Do as little as possible
Vectorise
Avoid copies

Do as little as possible

n <- 100000
cols <- 150
data_mat <- matrix(rnorm(n * cols, mean = 5), ncol = cols)
data <- as.data.frame(data_mat)

bench::mark(
  means <- colMeans(data_mat),
  means <- colMeans(data),
  check = F
)

# A tibble: 2 × 6
  expression                       min   median `itr/sec` mem_alloc `gc/sec`
  <bch:expr>                  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
1 means <- colMeans(data_mat)    437ms    437ms      2.29    25.4KB     0   
2 means <- colMeans(data)        456ms    458ms      2.18   114.5MB     2.18

Avoid copies

The code below samples 100 observations from a \(N(0, 1)\) distribution:

x <- c()
for(i in 1:100){
  x <- c(x, rnorm(1))
}

How could I make this code more efficient?

Avoid copies

loop_1 <- function(n){
  x <- c()
  for(i in 1:n){
    x <- c(x, rnorm(1))
  }
  return(x)
}

loop_2 <- function(n){
  x <- rep(NA, n)
  for(i in 1:n){
    x[i] <- rnorm(1)
  }
  return(x)
}

Avoid copies

bench::mark(
  loop_1(100),
  loop_2(100),
  check = F
)

# A tibble: 2 × 6
  expression       min   median `itr/sec` mem_alloc `gc/sec`
  <bch:expr>  <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
1 loop_1(100)    143µs    161µs     5552.     318KB     11.6
2 loop_2(100)    114µs    118µs     7529.     272KB     11.5

Avoid copies

bench::mark(
  loop_1(10000),
  loop_2(10000),
  check = F
)

# A tibble: 2 × 6
  expression         min   median `itr/sec` mem_alloc `gc/sec`
  <bch:expr>    <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
1 loop_1(10000)   85.7ms  104.8ms      9.22   406.3MB    20.0 
2 loop_2(10000)   11.8ms   12.2ms     70.6     24.5MB     9.80

Vectorise

The code below samples 100 observations from a \(N(0, 1)\) distribution:

x <- rep(NA, 100)
for(i in 1:100){
  x[i] <- rnorm(1)
}

How could I make this code more efficient?

Vectorise

for_loop_sample <- function(n){
  x <- rep(NA, n)
  for(i in 1:n){
    x[i] <- rnorm(1)
  }
}

bench::mark(
  x <- for_loop_sample(100),
  x <- rnorm(100),
  check=F
)

# A tibble: 2 × 6
  expression                     min   median `itr/sec` mem_alloc `gc/sec`
  <bch:expr>                <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
1 x <- for_loop_sample(100) 113.92µs 119.04µs     8069.  271.04KB     11.2
2 x <- rnorm(100)             5.42µs   6.33µs   151517.    3.32KB     15.2

Other options

Different data structures / algorithms
Parallelization
Rewrite code in C++

Class activity

https://sta279-f23.github.io/class_activities/ca_lecture_25.html