In today’s Programming Praxis we have to provide three different methods of calculating the ever-popular Fibonacci numbers; one exponential, one linear and one logarithmic. Let’s get started, shall we?
import Data.List import Criterion.Main
The naive exponential solution is trivial:
fibexp :: Int -> Integer fibexp 0 = 0 fibexp 1 = 1 fibexp n = fibexp (n - 1) + fibexp (n - 2)
For the linear method, we use the textbook lazy evaluation-based approach:
fiblin :: Int -> Integer fiblin n = fibs !! n where fibs = 0:1:zipWith (+) fibs (tail fibs)
The logarithmic solution requires two helper functions: the matrix multiplication function from a previous exercise and a way of raising a matrix to a power in log(n) time.
mult :: Num a => [[a]] -> [[a]] -> [[a]] mult a b = [map (sum . zipWith (*) r) $ transpose b | r <- a] matrixpower :: [[Integer]] -> Int -> [[Integer]] matrixpower m 1 = m matrixpower m n = (if even n then id else mult m) $ matrixpower (mult m m) (div n 2)
All that’s left to do to calculate Fibonacci numbers is raise the given matrix to the correct power and taking the lower-left element.
fiblog :: Int -> Integer fiblog 0 = 0 fiblog n = matrixpower [[1,1],[1,0]] n !! 1 !! 0
To benchmark the different solutions, we use the Criterion library.
main :: IO () main = defaultMain [bench "exp" $ nf fibexp 25 ,bench "lin" $ nf fiblin 25000 ,bench "log" $ nf fiblog 25000 ]
This gives the following timings: 174 ms for the exponential version, 69 ms for the linear one and 643 microseconds for the logarithmic solution, so we get a 100-fold speedup between the linear and logarithmic version at the cost of a factor of 6 increase in code size. Not a bad trade-off.