With Missings, Julia is slower than R

General Usage
4

Using BenchmarkTools (as suggested by Stefan) and Julia v0.7-alpha (as suggested by Chris), I see Julia being only slightly slower than R:

Edit: And the very latest Julia nightlies are even faster than R (see below)

julia> using BenchmarkTools

julia> @btime sum(skipmissing($y))
  1.329 s (6 allocations: 96 bytes)

> system.time(sum(x, na.rm = T))
   user  system elapsed 
  0.940   0.000   0.941

That’s actually pretty good, since presumably R’s sum() is just calling a C function. Calling a single (C-implemented) function on a vector of data is pretty much the best possible case for R/NumPy/MATLAB/etc.

The important difference is that Julia’s sum(skipmissing(y)) is just calling Julia code all the way down. So if you want to implement your own kind of summation, you actually can and it will still be fast. Compare:

julia> function mysum(x)
         result = zero(eltype(x))
         for element in x
           if !ismissing(element)
             result += element
           end
         end
         result
       end
mysum (generic function with 1 method)

julia> @btime mysum($y)
  809.128 ms (0 allocations: 0 bytes)

vs R:

> mysum <- function(x) {
+   result <- 0
+   for (element in x) {
+     if (!is.na(element)) {
+       result <- result + element
+     }
+   }
+  result
+ }
> system.time(mysum(x))
   user  system elapsed 
107.094   0.000 107.093

The hand-written R sum is over 100 times slower than Julia.

So yes, if all you ever want to do is compute sums over vectors, then R is fine. But if you want to do something that doesn’t have a fast built-in C implementation, then you really want Julia.

By the way,

Are sentinelle values to represent missing values really out of the question?

No, certainly not. Julia v0.7 stores vectors of things like Union{Float64, Missing} efficiently by using a sentinel bit for each element. Or you can implement your own, which is what DataArrays does: https://github.com/JuliaStats/DataArrays.jl/blob/49004b5f82bb92ec7790805a0dda10b8c3aeb68b/src/dataarray.jl#L32-L70 and it will be fast, too.

7 Likes