Implement a function to count frequency of each element in a list

sabu picture sabu · Nov 22, 2012 · Viewed 10.1k times · Source

I try to write a program which will count the frequency of each element in a list.

    In: "aabbcabb"
    Out: [("a",3),("b",4),("c",1)]

You can view my code in the following link: http://codepad.org/nyIECIT2 In this code the output of unique function would be like this

     In: "aabbcabb"
     Out: "abc"

Using the output of unique we wil count the frequency of the target list. You can see the code here also:

    frequencyOfElt xs=ans
       where ans=countElt(unique xs) xs
          unique []=[]
      unique xs=(head xs):(unique (filter((/=)(head xs))xs))
      countElt ref target=ans'
             where ans'=zip ref lengths
            lengths=map length $ zipWith($)(map[(=='a'),(==',b'),(==',c')](filter.(==))ref)(repeat target)

    Error:Syntax error in input (unexpected symbol "unique") 

But in ghci 6.13 other type of error are showing also

Few asked me what is the purpose of using [(=='a'),(==',b'),(==',c')]. What I expect: If ref="abc" and target="aabbaacc" then

    zipWith($) (map filter ref)(repeat target)

will show ["aaaa","bb","cc"] then I can use map length over this to get the frequency Here for filtering list according with the ref i use [(=='a'),(==',b'),(==',c')]

I assume some logical error lies [(=='a'),(==',b'),(==',c')] here..

Answer

Will Ness picture Will Ness · Nov 22, 2012

You didn't say whether you want to write it whole on your own, or whether it's OK to compose it from some standard functions.

import Data.List

g s = map (\x -> ([head x], length x)) . group . sort $ s

-- g = map (head &&& length) . group . sort     -- without the [...]

is the standard quick-n-dirty way to code it.


OK, so your original idea was to Code it Point-Free Style (certain tune playing in my head...):

frequencyOfElt :: (Eq a) => [a] -> [(a,Int)]
frequencyOfElt xs = countElt (unique xs) xs     -- change the result type
  where 
    unique [] = []
    unique (x:xs) = x : unique (filter (/= x) xs)  

    countElt ref target =   -- Code it Point-Free Style  (your original idea)
      zip 
        ref $               -- your original type would need (map (:[]) ref) here
        map length $
          zipWith ($)       -- ((filter . (==)) c) === (filter (== c))
            (zipWith ($) (repeat (filter . (==))) ref)  
            (repeat target)

I've changed the type here to the more reasonable [a] -> [(a,Int)] btw. Note, that

zipWith ($) fs (repeat z) === map ($ z) fs
zipWith ($) (repeat f) zs === map (f $) zs === map f zs

hence the code simplifies to

    countElt ref target =  
      zip 
        ref $              
        map length $
          map ($ target)      
            (zipWith ($) (repeat (filter . (==))) ref)  

and then

    countElt ref target =  
      zip 
        ref $              
        map length $
          map ($ target) $
            map (filter . (==)) ref

but map f $ map g xs === map (f.g) xs, so

    countElt ref target =  
      zip 
        ref $              
        map (length . ($ target) . filter . (==)) ref      -- (1)

which is a bit clearer (for my taste) written with a list comprehension,

    countElt ref target =  
        [ (c, (length . ($ target) . filter . (==)) c) | c <- ref] 
     == [ (c,  length ( ($ target) ( filter (== c))))  | c <- ref]     
     == [ (c,  length $ filter (== c) target)          | c <- ref]     

Which gives us an idea to re-write (1) further as

    countElt ref target =  
      zip <*> map (length . (`filter` target) . (==)) $ ref

but this obsession with point-free code becomes pointless here.


So going back to the readable list comprehensions, using a standard nub function which is equivalent to your unique, your idea becomes

import Data.List

frequencyOfElt xs = [ (c, length $ filter (== c) xs) | c <- nub xs]

This algorithm is actually quadratic (~ n^2), so it is worse than the first version above which is dominated by sort i.e. is linearithmic (~ n log(n)).


This code though can be manipulated further by a principle of equivalent transformations:

  = [ (c, length . filter (== c) $ sort xs) | c <- nub xs]

... because searching in a list is the same as searching in a list, sorted. Doing more work here -- will it pay off?..

  = [ (c, length . filter (== c) $ sort xs) | (c:_) <- group $ sort xs]

... right? But now, group had already grouped them by (==), so there's no need for the filter call to repeat the work already done by group:

  = [ (c, length . get c . group $ sort xs) | (c:_) <- group $ sort xs]
            where get c gs = fromJust . find ((== c).head) $ gs

  = [ (c, length g) | g@(c:_) <- group $ sort xs]

  = [ (head g, length g) | g <- group (sort xs)]

  = (map (head &&& length) . group . sort) xs

isn't it? And here it is, the same linearithmic algorithm from the beginning of this post, actually derived from your code by factoring out its hidden common computations, making them available for reuse and code simplification.