Welcome to the course
Optimizing R Code with Rcpp
Romain François
Consulting Datactive, ThinkR
R vs C++
R
- Great, flexible
- Slow
- Interpreted
C++
- Compiled
- More difficult
- Fast
Motivation
- Use Rcpp to make your code faster
- No need to know all of C++
- Focus on writing simple C++ functions
Course structure
- Introduction - basic C++ syntax
- C++ functions and control flow
- Vector classes
- Case studies
Measure performance
loopy version of max
slowmax <- function(x){
res <- x[1]
for ( i in 2:length(x) ){
if( x[i] > res ) res <- x[i]}
res }
Comparing performance with the microbenchmark
library(microbenchmark)
x <- rnorm(1e6)
microbenchmark( slowmax(x), max(x) )
Unit: milliseconds
expr min lq mean median uq max neval
slowmax(x) 31.649452 34.29454 36.344912 35.435299 37.188249 90.363038 100
max(x) 1.563559 1.74036 1.939367 1.847045 2.014684 3.340052 100
Evaluating simple C++ expressions with evalCpp
library(Rcpp)
evalCpp( "40 + 2" )
42
evalCpp( "exp(1.0)" )
2.718282
evalCpp( "sqrt(4.0)" )
2
Using
std::numeric_limits<int>::max()
to get the biggest representable 32 bit signed integer (int)
evalCpp(
"std::numeric_limits<int>::max()"
)
2147483647
( $\footnotesize \mathtt{2147483647 = 2^{31}-1}$)
Basic number types
C++ has rich set of number types
- Integer numbers:
int - Floating point numbers:
double
R
# Literal numbers are double
x <- 42
storage.mode(x)
"double"
# Integers need the L suffix y <- 42L storage.mode(y)z <- as.integer(42) storage.mode(z)
"integer"
"integer"
C++
# Suffix .0 forces a double
y <- evalCpp( "42.0" )
storage.mode(y)
"double"
library(Rcpp)
# Literal integers are int
x <- evalCpp( "42" )
storage.mode(x)
"integer"
Casting
Explicit casting with (double)
# Explicit cast
y <- evalCpp("(double)(40 + 2)")
storage.mode(y)
"double"
Beware of the integer division
# Integer division
evalCpp( "13 / 4" )
3
# Explicit cast, and hence use
# of double division
evalCpp( "(double)13 / 4" )
3.25
# Automatic conversion in R
13L / 4L
3.25
Let's practice!
Optimizing R Code with Rcpp
