Performance Haskell上数独求解器的优化

我用Haskell写了一个数独解算器。它遍历一个列表，当它找到“0”（一个空单元格）时，它将获得适合的数字并尝试：

import Data.List (group, (\\), sort)
import Data.Maybe (fromMaybe)

row :: Int -> [Int] -> [Int]
row y grid = foldl (\acc x -> (grid !! x):acc) [] [y*9 .. y*9+8]
    where y' = y*9
column :: Int -> [Int] -> [Int]
column x grid = foldl (\acc n -> (grid !! n):acc) [] [x,x+9..80]
box :: Int -> Int -> [Int] -> [Int]
box x y grid = foldl (\acc n -> (grid !! n):acc) [] [x+y*9*3+y' | y' <- [0,9,18], x <- [x'..x'+2]]
    where x' = x*3

isValid :: [Int] -> Bool
isValid grid = and [isValidRow, isValidCol, isValidBox]
    where isValidRow = isValidDiv row
          isValidCol = isValidDiv column
          isValidBox = and $ foldl (\acc (x,y) -> isValidList (box x y grid):acc) [] [(x,y) | x <- [0..2], y <- [0..2]]
          isValidDiv f = and $ foldl (\acc x -> isValidList (f x grid):acc) [] [0..8]
          isValidList = all (\x -> length x <= 1) . tail . group . sort -- tail removes entries that are '0'

isComplete :: [Int] -> Bool        
isComplete grid = length (filter (== 0) grid) == 0

solve :: Maybe [Int] -> Maybe [Int]
solve grid' = foldl f Nothing [0..80]
    where grid = fromMaybe [] grid' 
          f acc x
            | isValid grid = if isComplete grid then grid' else f' acc x
            | otherwise    = acc
          f' acc x 
            | (grid !! x) == 0 = case guess x grid of 
                Nothing -> acc
                Just x -> Just x
            | otherwise        = acc

guess :: Int -> [Int] -> Maybe [Int]
guess x grid
    | length valid /= 0 = foldl f Nothing valid
    | otherwise         = Nothing
    where valid = [1..9] \\ (row rowN grid ++ column colN grid ++ box (fst boxN) (snd boxN) grid) -- remove numbers already used in row/collumn/box
          rowN = x `div` 9 -- e.g. 0/9=0 75/9=8
          colN = x - (rowN * 9) -- e.g. 0-0=0 75-72=3
          boxN = (colN `div` 3, rowN `div` 3)
          before x = take x grid
          after x = drop (x+1) grid
          f acc y = case solve $ Just $ before x ++ [y] ++ after x of
            Nothing -> acc
            Just x -> Just x

在一秒钟内拍摄，但这一次：

sudoku :: [Int]
sudoku = [5,3,0,6,7,8,0,1,2,
          6,7,0,0,0,0,3,4,8,
          0,0,8,0,0,0,5,0,7,
          8,0,0,0,0,1,0,0,3,
          4,2,6,0,0,3,7,9,0,
          7,0,0,9,0,0,0,5,0,
          9,0,0,5,0,7,0,0,0,
          2,8,7,4,1,9,6,0,5,
          3,0,0,2,8,0,1,0,0]

sudoku :: [Int]
sudoku = [5,3,0,0,7,0,0,1,2,
          6,7,0,0,0,0,3,4,8,
          0,0,0,0,0,0,5,0,7,
          8,0,0,0,0,1,0,0,3,
          4,2,6,0,0,3,7,9,0,
          7,0,0,9,0,0,0,5,0,
          9,0,0,5,0,7,0,0,0,
          2,8,7,4,1,9,6,0,5,
          3,0,0,2,8,0,1,0,0]

我还没看完。我不认为这是该方法的问题，因为它确实返回正确的结果

---

分析表明，大部分时间都花在了“isValid”函数上。该功能是否有明显的低效/缓慢之处

isValidList=all（\x->length x实现当然是可以改进的，但这不是问题所在。问题是，对于第二个网格，简单的猜测和检查算法需要大量的回溯。即使将每个函数的速度提高1000倍，也会有网格需要宇宙年龄的几倍找到（第一个，如果网格不是唯一的）解决方案

isValidList = all (\x -> length x <= 1) . tail . group . sort -- tail removes entries that are '0'

你需要一个更好的算法来避免这种情况。避免这种情况的一个相当有效的方法是首先猜测可能性最小的平方。这并不能避免所有的坏情况，但可以大大减少它们

您还应该做的一件事是将
length thing==0
检查替换为
null thing
。在这里出现相对较短的列表时，效果是有限的，但通常情况下效果会很显著（通常也不应该使用
length list）