DEV Community

Myoungjin Jeon
Myoungjin Jeon

Posted on • Edited on • Originally published at jeongoon.github.io

Combinations in Haskell

#haskell #combinations #benchmark

Copyright (c) 2022 JEON Myoungjin

LICENSE: Open Software License 3.0

There was war between my favourite languages ...

Not So Precise Benchmark

And I am too obsessed with How to make combinations in haskell.

Combinations in Haskell Series

  1. Combinations In Haskell (called Tail After Tail)
  2. Tail After Tail Story (Combinations In Haskell)

Same Theory; Different Implementation

In programming world, the pseudo code or theory will trigger your programming implementation in many ways. This kind of diversion makes programming interesting as well.

Previously I made TailAfterTail.lhs

And I found that inits or tails are not quite necessary if I don't rely on scanl or zipWith.

It is probably generally accepted that some different kind of implementation, which consist of even small amount of codes, but which run at very high frequency, will eventually show huge gaps in performance after many iterations of execution.

So, this is about what I found during implementation of Tail After Tail combinations.

Module Begins

Firstly, I going to write down the all function will be exposed.

{-# LANGUAGE BangPatterns #-}

module TailAfterTail
  ( combinationsWithScanl
  , combinationsWithSingleStep
  , combinations
  , combinationsWithTwoSteps
  , allCombinationsWithScanl
  , allCombinationsWithSingleStep
  , allCombinationsWithTwoSteps
  , allCombinations
  ) where

import Data.List (tails, inits, scanl') -- only required for **WithScanl
Enter fullscreen mode Exit fullscreen mode

Original Version (scanl)

Please Find out more information HERE.

However, combinations1', flatten_allCombinationsGrouped and
genPart will be common helper functions.

combinations1' :: [a] -> [[[a]]]
combinations1' ms = [ [[m]] | m <- ms ]

flatten_allCombinationsGrouped allComboFunc = map concat . allComboFunc

genPart :: Foldable t => a -> t [[a]] -> [[a]]
genPart leader followerGroups = [ leader : followers
                                  | followers <- concat followerGroups ]
Enter fullscreen mode Exit fullscreen mode

And I define some helper functions.

usefulTails :: [a] -> [[a]]
usefulTails = init . tails

genStep :: [[[a]]] -> [a] -> [[[a]]]
genStep prevTails members' =
  zipWith genPart members' (usefulTails prevTails')
  where
    prevTails' = tail prevTails -- head is not useful

membersTails = reverse . tail . inits -- tail is used to skip empty list.
Enter fullscreen mode Exit fullscreen mode

and finally combinationsWithScanl family goes below.

allCombinationsWithScanl' :: [a] -> [[[[a]]]]
allCombinationsWithScanl' ms =
  scanl' genStep (combinations1' ms) (membersTails ms)

allCombinationsWithScanlGrouped :: [a] -> [[[a]]]
allCombinationsWithScanlGrouped =
  flatten_allCombinationsGrouped allCombinationsWithScanl'

allCombinationsWithScanl :: [a] -> [[a]]
allCombinationsWithScanl = concat . allCombinationsWithScanlGrouped
Enter fullscreen mode Exit fullscreen mode

Pure Implementation Without Scanl (SingleStep)

The following code is created without scanl or zipWith.

It gains slightly more performance with (bang pattern: !).
Which will be covered may be in another article.
But IMHO, it helps to reduce laziness and use less stack.

unsafe_allCombinationsWithSingleStep :: [a] -> [[[[a]]]]
unsafe_allCombinationsWithSingleStep members =
  let
    helper ! cases = -- bang pattern added
      let
        genStep (m:ms) (_:cs:[]) = [ [ m : c | c <- cs ] ]
        genStep (m:ms) (_:cs) =
          -- note       ^ : we don't use first element
          genPart m cs : genStep ms cs
      in
         cases : helper (genStep members cases)
  in
    helper . combinations1' $ members
Enter fullscreen mode Exit fullscreen mode

As you can see helper function is just an entry level wrapper function and make a recursion call.

genStep will actually create next cases and act as thunk which is evaluated later thanks to laziness in haskell.

I named the function as unsafe_ on purpose. Because helper function actually doesn't know when it will stop, and if you run unsafe_allCombinationsWithSingleStep in bare context will explode with exception.

sh> ghci
GHCi, version 8.10.7: https://www.haskell.org/ghc/  :? for help
Loaded GHCi configuration from /your/home/.config/ghc/ghci.conf
λ> :l 2022-04-15-Combinations-TailAfterTail'.lhs
[1 of 1] Compiling TailAfterTail    ( 2022-04-15-Combinations-TailAfterTail'.lhs, interpreted )
Ok, one module loaded.
λ> unsafe_allCombinationsWithSingleStep [1..5]
[[[[1]],[[2]],[[3]],[[4]],[[5]]],[[[1,2],[1,3],[1,4],[1,5]],[[2,3],[2,4],[2,5]],[[3,4],[3,5]],[[4,5]]],[[[1,2,3],[1,2,4],[1,2,5],[1,3,4],[1,3,5],[1,4,5]],[[2,3,4],[2,3,5],[2,4,5]],[[3,4,5]]],[[[1,2,3,4],[1,2,3,5],[1,2,4,5],[1,3,4,5]],[[2,3,4,5]]],[[[1,2,3,4,5]]],[[]*** Exception: 2022-04-15-Combinations-TailAfterTail'.lhs:(120,9)-(122,52): Non-exhaustive patterns in function genStep
Enter fullscreen mode Exit fullscreen mode

unsafe_allCombinationsWithSingleStepGrouped flatten the result but yet grouped by selection size, this is still unsafe but combinationsWith will handle it.

unsafe_allCombinationsWithSingleStepGrouped :: [a] -> [[[a]]]
unsafe_allCombinationsWithSingleStepGrouped =
  flatten_allCombinationsGrouped unsafe_allCombinationsWithSingleStep
Enter fullscreen mode Exit fullscreen mode

So now we could get all combinations by flatten a more time.

allCombinationsWithSingleStep :: [a] -> [[a]]
allCombinationsWithSingleStep members =
  concat
  --  this makes unsafe_* safe by limiting the size of list.
  . take (length members)
  . unsafe_allCombinationsWithSingleStepGrouped
  $ members
Enter fullscreen mode Exit fullscreen mode

With Two Steps

This is another version of without scanl. the Main improvement is that
this function separates the jobs into two operations:

  • create first cases from the previous tails.
  • create rest of cases and start next next case based on the result. 
allCombinationsWithTwoSteps' :: [a] -> [[[[a]]]]
allCombinationsWithTwoSteps'
  members@(fm:rms) = -- ^ fm : first member; rms: rest members
  let
    initFirstCase = [[fm]]
    initRestCases = combinations1' rms

    genFirstCases = genPart fm

    genRestCases _ [] = []
    genRestCases (m:ms) rcs@(_:rcs') = -- ^ rcs : rest of cases
      (genPart m $ rcs) : (genRestCases ms rcs')
Enter fullscreen mode Exit fullscreen mode

It looks almost identical when comparing to SingleStep but
now helper function knows exactly where to start as newTail is
memorized at the moment. It only saves time to tail by pattern matching
in SingleStep but resuts are propagated when the choices are growing.

BTW, tail by pattern matching means (_:cs) in the following code.

        genStep (m:ms) (_:cs) =
          -- note       ^ : we don't use first element
          [ m : c | c <- concat cs ] : genStep ms cs
Enter fullscreen mode Exit fullscreen mode

And let's wire them up with initial case and helper function

    helper [] = []
    helper ! prevTail =
      let
        newTail = genRestCases rms (tail prevTail)
      in
        ((genFirstCases prevTail) : newTail) : helper newTail
  in (initFirstCase : initRestCases) : helper initRestCases
Enter fullscreen mode Exit fullscreen mode

the following steps are similar to the other implementation.

allCombinationsWithTwoStepsGrouped :: [a] -> [[[a]]]
allCombinationsWithTwoStepsGrouped =
  flatten_allCombinationsGrouped allCombinationsWithTwoSteps'

allCombinationsWithTwoSteps :: [a] -> [[a]]
allCombinationsWithTwoSteps members =
  concat . allCombinationsWithTwoStepsGrouped $ members
Enter fullscreen mode Exit fullscreen mode

Another benefit of the TwoSteps implementation is that we can stop
easily because now newTail is always available and we could know whether
next step is available or not. I don't need to name it unsafe_ any more.

combinations variant from each implementation

Now, it's time to make select K out of given choice.

And I found that this is a common helper function:

combinationsWith 

combinationsWith :: ([a] -> [[[a]]]) -> [a] -> Int -> Int -> [[a]]
combinationsWith allComboGroupedFunc ms n1@selectFrom n2@selectTo =
  let
    ( isFlipped, n1', n2' ) = -- smaller value first
      if n1 < n2 then ( False
                      , max n1 0
                      , max n2 0)
      else            ( True
                      , max n2 0
                      , max n1 0)
      -- and ensure all range value are zero or positive by using `max`
    rangeLength = n2' - n1' + 1
    reverseIfNeeded
      | isFlipped = reverse
      | otherwise = id

  in
    -- note: read from the bottom
    concat                      -- 4. final flattening
    . reverseIfNeeded           -- 3. if user put opposite way, reverse it.
    . take rangeLength          -- 2. takes only interested lists
    . drop (pred n1')           -- 1. ignore some
    $ allComboGroupedFunc ms
Enter fullscreen mode Exit fullscreen mode

And all variant combinations* function available as below:

combinationsWithScanl      = combinationsWith allCombinationsWithScanlGrouped
combinationsWithSingleStep = combinationsWith unsafe_allCombinationsWithSingleStepGrouped
combinationsWithTwoSteps   = combinationsWith allCombinationsWithTwoStepsGrouped
Enter fullscreen mode Exit fullscreen mode

Benchmark

you can find the benchmark code on my github repository.
To save your time, THIS
is one of my benchmark result.

Choose Default allCombinations and combinations

After benchmarking, I found AllcombinationsWithTwoSteps shows best result
in all categories(small, medium, large) among them.

allCombinations = allCombinationsWithTwoSteps
combinations = combinationsWithTwoSteps
Enter fullscreen mode Exit fullscreen mode

Top comments (0)

Read next

sarahthomasdev profile image

Which iOS Performance Testing Tools should be used in 2025?

Sarah Thomas -

tuttelikz profile image

Creating C++ DLL and using it in .NET

San Askaruly -

farhan_ahmedfarhan41230 profile image

how can I add search functionality in this code?

Farhan Ahmed (Farhan41230301394) -

vania_gomes profile image

Entendendo props e useState no React: Como Passar e Gerenciar Dados Entre Componentes

Vânia Gomes -