RL.MC

Documentation

data MC_Opts

Constructors

MC_Opts

Fields

• o_alpha :: MC_Number
• o_maxlen :: Int
• o_maxlen_reward :: MC_Number

Instances

Show MC_Opts

Methods

showsPrec :: Int -> MC_Opts -> ShowS

show :: MC_Opts -> String

showList :: [MC_Opts] -> ShowS

defaultOpts :: MC_Opts

type MC_Number = Double

type Q s a = M s a MC_Number

type V s = HashMap s MC_Number

emptyQ :: MC_Number -> Q s a

q2v :: (Bounded a, Enum a, Eq a, Hashable a, Eq s, Hashable s) => Q s a -> V s

diffV :: (Eq s, Hashable s) => V s -> V s -> MC_Number

toV :: (Bounded a, Enum a, Eq a, Hashable a, Eq s, Hashable s) => Q s a -> V s

class (Fractional num, Ord s, Ord a, Show s, Show a, Bounded a, Enum a) => MC_Problem pr s a num | pr -> s, pr -> a, pr -> num where

Minimal complete definition

mc_is_terminal, mc_reward

Methods

mc_is_terminal :: pr -> s -> Bool

mc_reward :: pr -> s -> a -> s -> num

queryQ :: (Hashable s, Hashable k, MonadState (M s k v) f, Eq s, Eq k, Enum k, Bounded k) => s -> f [(k, v)]

modifyQ :: (Hashable a, Hashable s, MonadState (M s a num) m, Eq a, Eq s, Enum a, Bounded a) => s -> a -> (num -> num) -> m ()

data MC pr m s a

Constructors

MC

Fields

• mc_pr :: pr
• mc_transition :: s -> a -> m s

mc_es_learn :: (Monad m, Hashable s, Hashable a, MC_Problem pr s a MC_Number) => MC_Opts -> Q s a -> s -> a -> MC pr m s a -> m (Q s a)

MC-ES learning algorithm, pg 5.4. Alpha-learing rate is used instead of total averaging, maximum episode length is limited to make sure policy it terminates