Saturday, February 24, 2007

My first monadic program!

Problem 32 from Project Euler reads

The product 7254 is unusual, as the identity, 39 x 186 = 7254, containing multiplicand, multiplier, and product is 1 through 9 pandigital.

Find the sum of all products whose multiplicand/multiplier/product identity can be written as a 1 through 9 pandigital.

Not one of my brighter moments, the first approach I considered was applying goon-force to permute a list of eleven characters ('1' .. '9', 'x', and '=') -- a search space of almost 40 million -- looking for well-formed and valid statements.

Initially, I looked at using Text.Regex. As I was fishing for examples to crib from, I saw a suggestion that people may as well use Parsec, the monadic parser combinator library. So let's knock out the preliminaries:

> module Main where
> import Data.List hiding (map)
> import Text.ParserCombinators.Parsec

A parser for the simple expressions reads easily:

> num :: Parser Int
> num = do ns <- many1 digit
>          return $ read ns
> expression :: Parser (Int,Int,Int)
> expression = do multiplier <- num
>                 times <- char 'x'
>                 multiplicand <- num
>                 equals <- char '='
>                 product <- num
>                 return (multiplier, multiplicand, product)

Generating permutations is no problem in Haskell:

> permutations :: [a] -> [[a]]
> permutations [x] = [[x]]
> permutations xs =
>   [ y : zs
>   | (y,ys) <- selections xs
>   , zs     <- permutations ys
>   ]
> selections []     = []
> selections (x:xs) = (x,xs) : [(y,x:ys) | (y,ys) <- selections xs]

I'd been wanting to solve as a learning example one of the Project Euler problems using monads. As I tried to shoehorn the problem into a monadic solution, I remembered the characterization of the Maybe monad as being useful for computations that can fail, and I saw two possibilities for failure: garbage input (e.g., "x=123456789") and false statements (e.g., "1x2=3456789").

Having a particular permutation, check tests whether it's well-formed and valid:

check :: String -> Maybe Int
check s = do
  result <- parseExpr s
  p <- validProduct result
  return p

parseExpr :: String -> Maybe (Int,Int,Int)
parseExpr s =
  case parse expression "expression" s of
    Left err -> Nothing
    Right tuple@(mr,md,pr) -> return tuple

validProduct (mr, md, pr)
  | mr * md == pr = Just pr
  | otherwise = Nothing

The test goes just as you would describe it to someone else: parse the input to extract the components of the mulitiplication and then check whether the multiplication holds. Simple.

You wonder, 'But what about when the parse fails or when the statement is bogus?' Those checks are still happening, but the Maybe monad and the do-notation syntactic sugar are performing those checks implicitly! "Control" (to borrow an imperative concept) reaches the return p line only if both the parse and validProduct succeed. Otherwise, check bails and returns Nothing.

All that's left to do is feed it input and sum the result. Note: it's very s-l-o-w.

p32 = sum $ elems $ fromList $ catMaybes $ map check (permutations cs)
  where cs = ['1','2','3','4','5','6','7','8','9','x','=']

An obvious improvement is to use nub instead of (elems . fromList). Even better, lists are also monads, so Maybe distractions disappear with only a few very minor changes:

> p32 = sum (nub $ concatMap check (permutations cs))
>   where cs = ['1','2','3','4','5','6','7','8','9','x','=']
> main = print p32
> check :: (Monad m) => String -> m Int
> check s = do
>   result <- parseExpr s
>   p <- validProduct result
>   return p
> parseExpr :: (Monad m) => String -> m (Int,Int,Int)
> parseExpr s =
>     case parse expression "expression" s of
>       Left err -> fail (show err)
>       Right tuple@(mr,md,pr) -> return tuple
> validProduct :: (Monad m) => (Int,Int,Int) -> m Int
> validProduct (mr, md, pr)
>   | mr * md == pr = return pr
>   | otherwise = fail "invalid product"

This works because failure in the list monad is the empty list, and concatMap gets rid of all the empties.

Saturday, February 17, 2007

Very touching short animation

I saw this video when my wife was in St. Louis at a baby shower for one of the girls she played volleyball with in college and my kids were at my mother-in-law's. So moving.