haskell - State Monadic 函数内的 runState 不起作用

Question

我正在尝试解决“AI - 现代方法”一书中的问题 2.8，该问题涉及单元格网格并选择随机移动来导航网格。

2.7 Implement an environment for a n X m rectangular room, where each square has a 5% chance of containing dirt, and n and m are chosen at random from the range 8 to 15, inclusive.

2.8 Design and implement a pure reflex agent for the environment of Exercise 2.7, ignoring the requirement of returning home, and measure its performance.

所以我使用了两个状态单子(monad) - 一个以 Grid 作为状态，另一个以 StdGen 作为状态。代码编译时没有任何错误，但当我从 GHCi 运行它时，它被卡住并且不返回。

代码的相关部分:

支持代码

type RandomState = State StdGen

makeGrid :: (Int, Int) -> (Int, Int) -> Float -> RandomState Grid

doAction :: Action -> Cleaner -> State Grid Cleaner

getRandomR :: Random a => (a, a) -> RandomState a
getRandomR limits = do
  gen <- get
  let (val, gen') = randomR limits gen
  put gen'
  return val

chooseAction :: Percepts -> RandomState Action
chooseAction percepts
  | PhotoSensor `elem` percepts = return SuckDirt
  | InfraredSensor `elem` percepts = return TurnOff
  | TouchSensor `elem` percepts = return TurnLeft
  | otherwise = do
    r <- getRandomR ((1, 3) :: (Int, Int))
    case r of
      1 -> return GoForward
      2 -> return TurnRight
      3 -> return TurnLeft

主要代码

runCleaner :: Int -> Cleaner -> StateT Grid RandomState Cleaner
runCleaner turnsLeft cleaner@(Cleaner _ _ _ ph _) =
  if turnsLeft == 0
    then return cleaner
    else do
      grid <- get
      gen <- lift $ get
      cleaner <- case ph of
        [] -> do
          let (cleaner, grid) = runState (doAction GoForward cleaner) grid
          put grid
          return cleaner
        _ -> do
          let (action, gen) = runState (chooseAction (head ph)) gen
          lift $ put gen

          let (cleaner, grid) = runState (doAction action cleaner) grid
          put grid
          return cleaner

      case clState cleaner of
        Off -> return cleaner
        On -> runCleaner (turnsLeft - 1) cleaner

simulateOnGrid :: Int -> Grid -> StdGen -> (Cleaner, Grid)
simulateOnGrid maxTurns grid gen = 
  evalState (runStateT (runCleaner maxTurns cleaner) grid) gen
  where cleaner = createCleaner (fromJust $ cell (0,0) grid) East

我从 GHCi 调用 simulateOnGrid 函数，如下所示:

> gen <- newStdGen
> let grid = evalState (makeGrid (8,15) (8,15) 0.05) gen
> simulateOnGrid 5 grid gen

并且代码卡在了以下行:

let (cleaner, grid) = runState (doAction GoForward cleaner) grid

我已经通过在代码中添加痕迹来确认这一点。对 doAction 函数的调用永远不会发生。

问题似乎是在 runCleaner 函数中使用了 runState，但我找不到任何原因。

请说明原因以及是否有办法解决此问题。

此外，在单子(monad)函数中使用 runState 对我来说感觉不对。如果有更好的方法请提出。

Answer 1

在 let 绑定(bind)的右侧，被绑定(bind)的名称在范围内，因此当您编写时

let (cleaner, grid) = runState (doAction GoForward cleaner) grid

=右侧的cleaner和grid与左侧的相同。当您将操作的输出作为其输入反馈时，这可能会导致无限循环!为了避免这种情况，请为输出使用不同的名称。

let (cleaner', grid') = runState (doAction GoForward cleaner) grid

<小时/>

除此之外，您说这样使用 runState 很奇怪，这是绝对正确的。我认为如果将 doAction 的类型更改为

，您可以大大简化事情

doAction :: Monad m => Action -> Cleaner -> StateT Grid m Cleaner

您没有提供此函数的主体，但我猜测它仍然可以使用这种约束较少的类型签名。

现在您不必再手动获取和放置状态了，因为 doAction 可以直接在您的 monad 中运行，并且 chooseAction 可以运行首先举起它。使用它，您的 case 表达式可以写得更加简洁:

cleaner <- case ph of
    [] -> doAction GoForward cleaner
    _  -> do action <- lift $ chooseAction (head ph)
             doAction action cleaner