Safe Haskell	None
Language	Haskell2010

ALG.TD3

Description

Twin Delayed Deep Deterministic Policy Gradient Algorithm

Synopsis

data PolicyNet = PolicyNet {
- pLayer0 :: Linear
- pLayer1 :: Linear
- pLayer2 :: Linear
}
data CriticNet = CriticNet {
- q1Layer0 :: Linear
- q1Layer1 :: Linear
- q1Layer2 :: Linear
- q2Layer0 :: Linear
- q2Layer1 :: Linear
- q2Layer2 :: Linear
}
data Agent = Agent {
- φ :: PolicyNet
- φ' :: PolicyNet
- θ :: CriticNet
- θ' :: CriticNet
- φOptim :: Adam
- θOptim :: Adam
- actLo :: Float
- actHi :: Float
- σA :: Tensor
- σE :: Tensor
- c' :: Float
}
mkAgent :: Params -> Int -> Int -> IO Agent
saveAgent' :: FilePath -> Agent -> IO ()

Documentation

data PolicyNet Source #

Actor Network Architecture

Constructors

PolicyNet
Fields pLayer0 :: Linear pLayer1 :: Linear pLayer2 :: Linear

Instances

Instances details

Show PolicyNet Source #
Instance details Defined in ALG.TD3 Methods showsPrec :: Int -> PolicyNet -> ShowS # show :: PolicyNet -> String # showList :: [PolicyNet] -> ShowS #
Generic PolicyNet Source #
Instance details Defined in ALG.TD3 Associated Types type Rep PolicyNet :: Type -> Type # Methods from :: PolicyNet -> Rep PolicyNet x # to :: Rep PolicyNet x -> PolicyNet #
Parameterized PolicyNet Source #
Instance details Defined in ALG.TD3 Methods flattenParameters :: PolicyNet -> [Parameter] # _replaceParameters :: PolicyNet -> ParamStream PolicyNet #
type Rep PolicyNet Source #
Instance details Defined in ALG.TD3 type Rep PolicyNet = D1 ('MetaData "PolicyNet" "ALG.TD3" "acid-0.1.0.0-Kasfqr97FKi1MHFsFLmEZd" 'False) (C1 ('MetaCons "PolicyNet" 'PrefixI 'True) (S1 ('MetaSel ('Just "pLayer0") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Linear) :: (S1 ('MetaSel ('Just "pLayer1") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Linear) :: S1 ('MetaSel ('Just "pLayer2") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Linear))))

data CriticNet Source #

Critic Network Architecture

Constructors

CriticNet
Fields q1Layer0 :: Linear q1Layer1 :: Linear q1Layer2 :: Linear q2Layer0 :: Linear q2Layer1 :: Linear q2Layer2 :: Linear

Instances

Instances details

Show CriticNet Source #
Instance details Defined in ALG.TD3 Methods showsPrec :: Int -> CriticNet -> ShowS # show :: CriticNet -> String # showList :: [CriticNet] -> ShowS #
Generic CriticNet Source #
Instance details Defined in ALG.TD3 Associated Types type Rep CriticNet :: Type -> Type # Methods from :: CriticNet -> Rep CriticNet x # to :: Rep CriticNet x -> CriticNet #
Parameterized CriticNet Source #
Instance details Defined in ALG.TD3 Methods flattenParameters :: CriticNet -> [Parameter] # _replaceParameters :: CriticNet -> ParamStream CriticNet #
type Rep CriticNet Source #
Instance details Defined in ALG.TD3 type Rep CriticNet = D1 ('MetaData "CriticNet" "ALG.TD3" "acid-0.1.0.0-Kasfqr97FKi1MHFsFLmEZd" 'False) (C1 ('MetaCons "CriticNet" 'PrefixI 'True) ((S1 ('MetaSel ('Just "q1Layer0") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Linear) :: (S1 ('MetaSel ('Just "q1Layer1") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Linear) :: S1 ('MetaSel ('Just "q1Layer2") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Linear))) :: (S1 ('MetaSel ('Just "q2Layer0") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Linear) :: (S1 ('MetaSel ('Just "q2Layer1") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Linear) :*: S1 ('MetaSel ('Just "q2Layer2") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Linear)))))

data Agent Source #

TD3 Agent

Constructors

Agent
Fields φ :: PolicyNet Online Policy φ φ' :: PolicyNet Target Policy φ' θ :: CriticNet Online Critic θ θ' :: CriticNet Target Critic θ' φOptim :: Adam Policy Optimizer θOptim :: Adam Critic Optimizer actLo :: Float Lower bound of Action space actHi :: Float Upper bound of Action space σA :: Tensor Action Noise σE :: Tensor Eval Noise c' :: Float Noise Clipping

Instances

Instances details

Show Agent Source #
Instance details Defined in ALG.TD3 Methods showsPrec :: Int -> Agent -> ShowS # show :: Agent -> String # showList :: [Agent] -> ShowS #
Generic Agent Source #
Instance details Defined in ALG.TD3 Associated Types type Rep Agent :: Type -> Type # Methods from :: Agent -> Rep Agent x # to :: Rep Agent x -> Agent #
Agent Agent Source #	TD3 Agent is an implements Agent
Instance details Defined in ALG.TD3 Methods saveAgent :: FilePath -> Agent -> IO Agent Source # loadAgent :: Params -> FilePath -> Int -> Int -> Int -> IO Agent Source # act :: Agent -> Tensor -> IO Tensor Source # act' :: Agent -> Tensor -> IO Tensor Source # act'' :: Agent -> Tensor -> Tensor Source # updatePolicy :: Params -> CircusUrl -> Tracker -> Int -> [Transition] -> Agent -> IO Agent Source #
type Rep Agent Source #
Instance details Defined in ALG.TD3 type Rep Agent = D1 ('MetaData "Agent" "ALG.TD3" "acid-0.1.0.0-Kasfqr97FKi1MHFsFLmEZd" 'False) (C1 ('MetaCons "Agent" 'PrefixI 'True) (((S1 ('MetaSel ('Just "\966") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 PolicyNet) :: S1 ('MetaSel ('Just "\966'") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 PolicyNet)) :: (S1 ('MetaSel ('Just "\952") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 CriticNet) :: (S1 ('MetaSel ('Just "\952'") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 CriticNet) :: S1 ('MetaSel ('Just "\966Optim") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Adam)))) :: ((S1 ('MetaSel ('Just "\952Optim") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Adam) :: (S1 ('MetaSel ('Just "actLo") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Float) :: S1 ('MetaSel ('Just "actHi") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Float))) :: (S1 ('MetaSel ('Just "\963A") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Tensor) :: (S1 ('MetaSel ('Just "\963E") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Tensor) :: S1 ('MetaSel ('Just "c'") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 Float))))))

mkAgent :: Params -> Int -> Int -> IO Agent Source #

Agent constructor

saveAgent' :: FilePath -> Agent -> IO () Source #

Save an Agent and return the agent