When to use Haskell monads

Question

I\'m implementing a combinatorial optimization algorithm in Haskell:

Given an initial candidate solution, repeat until stopping criteria are met:

  1. Determine         


        

Answer 1

Here's a pseudocodey sketch of how you might use the State monad to thread the search state through the computation:

import Control.Monad.State

type SearchState = ...
type Move = ...
type Fitness = ...

determineMoves :: State SearchState [Move]
determineMoves = do
  -- since determineMoves is in the State monad, we can grab the state here
  st <- get
  ...

evaluateMoves :: [Move] -> [(Move, Fitness)]
evaluateMoves = ...

chooseMove :: [(Move, Fitness)] -> Move
chooseMove = ...

-- makeMove is not itself monadic, but operates on the SearchState
-- type we're threading through with the State monad
makeMove :: Move -> SearchState -> SearchState
makeMove m st = ...

loop :: State SearchState ()
loop = do
  moves <- determineMoves
  let candidates = evaluateMoves moves
      move = chooseMove candidates
  -- we pass a function (SearchState -> SearchState) to modify in 
  -- order to update the threaded SearchState
  modify (makeMove move)
  loop

Notice that even though your main computation is in the state monad, not every component has to be in the monad. Here, evaluateMoves and chooseMove are non-monadic, and I've used let to show you how to explicitly integrate them into a do block. Once you get comfortable with this style, though, you'll probably want to get comfortable using <$> (aka fmap) and function composition to get more succinct:

loop :: State SearchState ()
loop = do
  move <- (chooseMove . evaluateMoves) <$> determineMoves
  modify (makeMove move)
  loop