Difference between revisions of "Random shuffle"

From HaskellWiki
Jump to navigation Jump to search
m (removed need for scoped variables, etc)
(Fix bug in MonadRandom shuffle (http://mail.haskell.org/pipermail/haskell-cafe/2017-May/127040.html))
 
(5 intermediate revisions by 3 users not shown)
Line 2: Line 2:
   
 
Shuffling a list, i.e. creating a random permutation, is not easy to do correctly. Each permutation should have the same probability.
 
Shuffling a list, i.e. creating a random permutation, is not easy to do correctly. Each permutation should have the same probability.
  +
  +
== Packages ==
  +
  +
There are ready made packages available from Hackage:
  +
* [https://hackage.haskell.org/package/shuffle shuffle]
  +
* [https://hackage.haskell.org/package/random-shuffle random-shuffle]
   
 
== Imperative algorithm ==
 
== Imperative algorithm ==
Line 70: Line 76:
   
 
This is a lot simpler than the purely functional algorithm linked below.
 
This is a lot simpler than the purely functional algorithm linked below.
  +
  +
Here's a variation using the MonadRandom package:
  +
  +
<haskell>
  +
import Control.Monad
  +
import Control.Monad.ST
  +
import Control.Monad.Random
  +
import System.Random
  +
import Data.Array.ST
  +
import GHC.Arr
  +
  +
shuffle :: RandomGen g => [a] -> Rand g [a]
  +
shuffle xs = do
  +
let l = length xs
  +
rands <- forM [0..(l-2)] $ \i -> getRandomR (i, l-1)
  +
let ar = runSTArray $ do
  +
ar <- thawSTArray $ listArray (0, l-1) xs
  +
forM_ (zip [0..] rands) $ \(i, j) -> do
  +
vi <- readSTArray ar i
  +
vj <- readSTArray ar j
  +
writeSTArray ar j vi
  +
writeSTArray ar i vj
  +
return ar
  +
return (elems ar)
  +
  +
*Main> evalRandIO (shuffle [1..10])
  +
[6,5,1,7,10,4,9,2,8,3]
  +
</haskell>
   
 
== Other implemenations ==
 
== Other implemenations ==
   
 
=== Purely functional ===
 
=== Purely functional ===
  +
* Using Data.Map, O(n * log n)
  +
<haskell>
  +
import System.Random
  +
import Data.Map
  +
  +
fisherYatesStep :: RandomGen g => (Map Int a, g) -> (Int, a) -> (Map Int a, g)
  +
fisherYatesStep (m, gen) (i, x) = ((insert j x . insert i (m ! j)) m, gen')
  +
where
  +
(j, gen') = randomR (0, i) gen
  +
  +
fisherYates :: RandomGen g => g -> [a] -> ([a], g)
  +
fisherYates gen [] = ([], gen)
  +
fisherYates gen l =
  +
toElems $ foldl fisherYatesStep (initial (head l) gen) (numerate (tail l))
  +
where
  +
toElems (x, y) = (elems x, y)
  +
numerate = zip [1..]
  +
initial x gen = (singleton 0 x, gen)
  +
</haskell>
 
* [http://okmij.org/ftp/Haskell/perfect-shuffle.txt Purely functional O(n log n) random shuffle algorithm].
 
* [http://okmij.org/ftp/Haskell/perfect-shuffle.txt Purely functional O(n log n) random shuffle algorithm].
   
Line 120: Line 173:
 
["cbf","dec","edb","fae","bda","cde"]
 
["cbf","dec","edb","fae","bda","cde"]
 
</haskell>
 
</haskell>
  +
  +
[[Category:code]]

Latest revision as of 13:10, 9 May 2017

The problem

Shuffling a list, i.e. creating a random permutation, is not easy to do correctly. Each permutation should have the same probability.

Packages

There are ready made packages available from Hackage:

Imperative algorithm

The standard imperative algorithm can be implemented as follows:

import System.Random
import Data.Array.IO
import Control.Monad

-- | Randomly shuffle a list
--   /O(N)/
shuffle :: [a] -> IO [a]
shuffle xs = do
        ar <- newArray n xs
        forM [1..n] $ \i -> do
            j <- randomRIO (i,n)
            vi <- readArray ar i
            vj <- readArray ar j
            writeArray ar j vi
            return vj
  where
    n = length xs
    newArray :: Int -> [a] -> IO (IOArray Int a)
    newArray n xs =  newListArray (1,n) xs

Or one can use ST to avoid needing IO:

import System.Random
import Data.Array.ST
import Control.Monad
import Control.Monad.ST
import Data.STRef

-- | Randomly shuffle a list without the IO Monad
--   /O(N)/
shuffle' :: [a] -> StdGen -> ([a],StdGen)
shuffle' xs gen = runST (do
        g <- newSTRef gen
        let randomRST lohi = do
              (a,s') <- liftM (randomR lohi) (readSTRef g)
              writeSTRef g s'
              return a
        ar <- newArray n xs
        xs' <- forM [1..n] $ \i -> do
                j <- randomRST (i,n)
                vi <- readArray ar i
                vj <- readArray ar j
                writeArray ar j vi
                return vj
        gen' <- readSTRef g
        return (xs',gen'))
  where
    n = length xs
    newArray :: Int -> [a] -> ST s (STArray s Int a)
    newArray n xs =  newListArray (1,n) xs

And if you are using IO's hidden StdGen you can wrap this as usual:

shuffleIO :: [a] -> IO [a]
shuffleIO xs = getStdRandom (shuffle' xs)

This is a lot simpler than the purely functional algorithm linked below.

Here's a variation using the MonadRandom package:

import Control.Monad
import Control.Monad.ST
import Control.Monad.Random
import System.Random
import Data.Array.ST
import GHC.Arr

shuffle :: RandomGen g => [a] -> Rand g [a]
shuffle xs = do
    let l = length xs
    rands <- forM [0..(l-2)] $ \i -> getRandomR (i, l-1)
    let ar = runSTArray $ do
        ar <- thawSTArray $ listArray (0, l-1) xs
        forM_ (zip [0..] rands) $ \(i, j) -> do
            vi <- readSTArray ar i
            vj <- readSTArray ar j
            writeSTArray ar j vi
            writeSTArray ar i vj
        return ar
    return (elems ar)

*Main> evalRandIO (shuffle [1..10])
[6,5,1,7,10,4,9,2,8,3]

Other implemenations

Purely functional

  • Using Data.Map, O(n * log n)
import System.Random
import Data.Map

fisherYatesStep :: RandomGen g => (Map Int a, g) -> (Int, a) -> (Map Int a, g)
fisherYatesStep (m, gen) (i, x) = ((insert j x . insert i (m ! j)) m, gen')
  where
    (j, gen') = randomR (0, i) gen

fisherYates :: RandomGen g => g -> [a] -> ([a], g)
fisherYates gen [] = ([], gen)
fisherYates gen l = 
  toElems $ foldl fisherYatesStep (initial (head l) gen) (numerate (tail l))
  where
    toElems (x, y) = (elems x, y)
    numerate = zip [1..]
    initial x gen = (singleton 0 x, gen)

Drawing without replacement

  • uses New_monads/MonadRandom
  • allows you to not shuffle the entire list but only part of it (drawing elements without replacement)
  • allows you to take multiple drawings/shufflings at once, which can save some array building
{- | @grabble xs m n@ is /O(m*n')/, where @n' = min n (length xs)@
     Chooses @n@ elements from @xs@, without putting back,
     and that @m@ times. -}
grabble :: MonadRandom m => [a] -> Int -> Int -> m [[a]]
grabble xs m n = do
    swapss <- replicateM m $ forM [0 .. min (maxIx - 1) n] $ \i -> do
                j <- getRandomR (i, maxIx)
                return (i, j)
    return $ map (take n . swapElems xs) swapss
    where
        maxIx   = length xs - 1

grabbleOnce :: MonadRandom m => [a] -> Int -> m [a]
grabbleOnce xs n = head `liftM` grabble xs 1 n

swapElems  :: [a] -> [(Int, Int)] -> [a]
swapElems xs swaps = elems $ runSTArray (do
    arr <- newListArray (0, maxIx) xs
    mapM_ (swap arr) swaps
    return arr)
    where
        maxIx   = length xs - 1
        swap arr (i,j) = do
            vi <- readArray arr i
            vj <- readArray arr j
            writeArray arr i vj
            writeArray arr j vi

So e.g.

*Main MonadRandom Random> evalRand (grabble "abcdef" 6 3) (mkStdGen 0)
["fbd","efb","bef","adc","cef","eac"]
*Main MonadRandom Random> grabble "abcdef" 6 3
["fce","dfa","ebf","edb","cea","dbc"]
*Main MonadRandom Random> grabble "abcdef" 6 3
["cbf","dec","edb","fae","bda","cde"]