Test.QuickCheck不能正確生成隨機值

Question

我正在使用Test.QuickCheck生成隨機數獨謎題。

data Sudoku = Sudoku { getSudoku :: [[Maybe Int]] } deriving (Show, Eq) 

rows :: Sudoku -> [[Maybe Int]] 
rows (Sudoku rs) = rs 

--B1 
printSudoku :: Sudoku -> IO() 
printSudoku s = do 
    putStrLn . unlines . map (map (maybe '.' (head . show))) $ rows s 

--C1 
cell :: Gen (Maybe Int) 
cell = suchThatMaybe (frequency [(90, choose (0,0)),(10, choose(1,9))]) (/=0) 

--C2 
instance Arbitrary Sudoku where 
    arbitrary = do 
     rows <- sequence [ sequence [ cell | j <- [1..9] ] | i <- [1..9] ] 
     return (Sudoku rows) 

--C3 
prop_Sudoku :: Sudoku -> Bool 
prop_Sudoku = isSudoku 

checkRandomSudoku :: IO [Bool] 
checkRandomSudoku = do 
    s <- sample' (arbitrary :: Gen Sudoku) 
    return $ map isSudoku s 

該代碼運行良好。然而，當我執行

一個< - 樣本」（任意::創數獨）

序列$地圖printSudoku一個

返回是這樣的：

....5..3. 
...4..... 
...2..... 
...5..... 
......... 
...33.... 
...5..... 
...2.4... 
......... 

......... 
.343..... 
......... 
......9.2 
......... 
45....5.1 
.2......7 
.7..88.34 
.9....6.. 

....2..8. 
.2121638. 
.7.7...9. 
4..45.6.. 
.....6.2. 
..6.6.... 
53..9.6.. 
..9....7. 
.47892... 

.373411.. 
5...3282. 
...45..9. 
8989..18. 
31.8113.. 
9..35.6.. 
4.685.... 
.4....39. 
7..6.5.76 

48.178.53 
1.871.4.4 
3165.17.. 
.1...7.59 
.98126.51 
6.6...775 
9.4636952 
.5..239.. 
372.....8 

.34.73129 
.5.8.27.1 
344.34931 
28.6.94.1 
6327.3..8 
3743.5496 
93...7984 
..82.8... 
..3.54.93 

273847853 
5568.7465 
832.73515 
3766..6.7 
.7.196256 
1.96.9.3. 
.7156.268 
1615.196. 
.392..633 

731652284 
863.8.768 
31..5.5.6 
961.5.467 
1245.1159 
5..275471 
52.727759 
6.656.849 
99.72352. 

這顯然是根本不是隨機的。空細胞的分佈開始時非常高，然後緩慢下降。我是否使用了錯誤的功能或以錯誤的方式使用？謝謝

Answer 1

您看到這種行爲的原因是，QuickCheck會逐漸嘗試更大的測試用例。這在手冊的resizing部分有描述。

在suchThatMaybe的情況下，增加大小會使其重試生成一個任意匹配您傳遞給它的謂詞。你可以在源文件中看到http://hackage.haskell.org/package/QuickCheck-2.6/docs/src/Test-QuickCheck-Gen.html#suchThatMaybe。有趣的代碼是：

try _ 0 = return Nothing 
try k n = do x <- resize (2*k+n) gen 
      if p x then return (Just x) else try (k+1) (n-1) 

發生了什麼事就是n取決於傳遞給發電機的大小參數suchThatMaybe重試n倍。

sample' tries the sizes [0,2..20]它會一直傳播到suchThatMaybe。

您可以通過發電機呼籲resize覆蓋增加尺寸：

>>> a <- sample' $ (resize 2 arbitrary :: Gen Sudoku) 
>>> sequence $ map printSudoku a 
.....8.5. 
.......4. 
......... 
4..25...5 
......9.. 
....5.... 
........7 
.....7... 
...4..... 

...5..... 
.......4. 
...6...6. 
..14..... 
...7...7. 
....2.... 
.....6... 
...4..572 
.4.....6. 

..6..8... 
..4...... 
......... 
......9.. 
839...... 
67..4.... 
.5....... 
....5.... 
........3 

4...1.... 
7..9.39.. 
....6.... 
...4.1... 
......... 
......... 
..9....6. 
.2.9...84 
.....8... 

.64...... 
..3.44... 
4.......4 
....1..8. 
.9....... 
34....... 
.....6... 
18.2..593 
.4.7..... 

......... 
...8..6.. 
.2......5 
...5..... 
.2....... 
........6 
.3....13. 
8.1.2..85 
....5.... 

..7...... 
..67...5. 
..6...... 
27....1.9 
.9....... 
78.....7. 
......34. 
.......2. 
..81...81 

3.1...... 
......... 
.....6... 
......... 
.16.71... 
......... 
.2....... 
......... 
.....9.1. 

..65..6.9 
........5 
..1.4.... 
....86... 
.2..2..2. 
.....9... 
..6...... 
......... 
...7..855 

.......94 
...14..8. 
.....4... 
...3....9 
......... 
.....5... 
.5....... 
45.....8. 
..48..... 

4........ 
......3.. 
5......4. 
.4..6..2. 
..3...... 
......... 
..9...... 
6..9..... 
....7....