Safe Haskell	None
Language	Haskell2010

TD3

Description

Twin Delayed Deep Deterministic Policy Gradient Algorithm

Synopsis

algorithm :: Algorithm
data ActorNetSpec = ActorNetSpec {
- pObsDim :: Int
- pActDim :: Int
}
data CriticNetSpec = CriticNetSpec {
- qObsDim :: Int
- qActDim :: Int
}
data ActorNet = ActorNet {
- 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 :: ActorNet
- φ' :: ActorNet
- θ :: CriticNet
- θ' :: CriticNet
- φOptim :: Adam
- θOptim :: Adam
}
mkAgent :: Int -> Int -> IO Agent
saveAgent :: String -> Agent -> IO ()
saveAgent' :: String -> Agent -> IO Agent
loadAgent :: String -> Int -> Int -> Int -> IO Agent
π :: ActorNet -> Tensor -> Tensor
q :: CriticNet -> Tensor -> Tensor -> (Tensor, Tensor)
q' :: CriticNet -> Tensor -> Tensor -> Tensor
act :: Agent -> Tensor -> IO Tensor
act' :: Int -> Agent -> Tensor -> IO Tensor
evaluate :: Agent -> Tensor -> IO Tensor
train :: Int -> Int -> HymURL -> TrackingURI -> IO Agent
updatePolicy :: Int -> Agent -> Tracker -> Buffer Tensor -> Int -> IO Agent

Documentation

algorithm :: Algorithm Source #

Algorithm ID

data ActorNetSpec Source #

Actor Network Specification

Constructors

ActorNetSpec
Fields pObsDim :: Int pActDim :: Int

Instances

Instances details

Eq ActorNetSpec Source #
Instance details Defined in TD3 Methods (==) :: ActorNetSpec -> ActorNetSpec -> Bool # (/=) :: ActorNetSpec -> ActorNetSpec -> Bool #
Show ActorNetSpec Source #
Instance details Defined in TD3 Methods showsPrec :: Int -> ActorNetSpec -> ShowS # show :: ActorNetSpec -> String # showList :: [ActorNetSpec] -> ShowS #
Randomizable ActorNetSpec ActorNet Source #	Actor Network Weight initialization
Instance details Defined in TD3 Methods sample :: ActorNetSpec -> IO ActorNet #

data CriticNetSpec Source #

Critic Network Specification

Constructors

CriticNetSpec
Fields qObsDim :: Int qActDim :: Int

Instances

Instances details

Eq CriticNetSpec Source #
Instance details Defined in TD3 Methods (==) :: CriticNetSpec -> CriticNetSpec -> Bool # (/=) :: CriticNetSpec -> CriticNetSpec -> Bool #
Show CriticNetSpec Source #
Instance details Defined in TD3 Methods showsPrec :: Int -> CriticNetSpec -> ShowS # show :: CriticNetSpec -> String # showList :: [CriticNetSpec] -> ShowS #
Randomizable CriticNetSpec CriticNet Source #	Critic Network Weight initialization
Instance details Defined in TD3 Methods sample :: CriticNetSpec -> IO CriticNet #

data ActorNet Source #

Actor Network Architecture

Constructors

ActorNet
Fields pLayer0 :: Linear pLayer1 :: Linear pLayer2 :: Linear

Instances

Instances details

Show ActorNet Source #
Instance details Defined in TD3 Methods showsPrec :: Int -> ActorNet -> ShowS # show :: ActorNet -> String # showList :: [ActorNet] -> ShowS #
Generic ActorNet Source #
Instance details Defined in TD3 Associated Types type Rep ActorNet :: Type -> Type # Methods from :: ActorNet -> Rep ActorNet x # to :: Rep ActorNet x -> ActorNet #
Parameterized ActorNet Source #
Instance details Defined in TD3 Methods flattenParameters :: ActorNet -> [Parameter] # _replaceParameters :: ActorNet -> ParamStream ActorNet #
Randomizable ActorNetSpec ActorNet Source #	Actor Network Weight initialization
Instance details Defined in TD3 Methods sample :: ActorNetSpec -> IO ActorNet #
type Rep ActorNet Source #
Instance details Defined in TD3 type Rep ActorNet = D1 ('MetaData "ActorNet" "TD3" "edelwace-0.1.0.0-NlkYKOSwr4E6F4MQQAmdJ" 'False) (C1 ('MetaCons "ActorNet" '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 TD3 Methods showsPrec :: Int -> CriticNet -> ShowS # show :: CriticNet -> String # showList :: [CriticNet] -> ShowS #
Generic CriticNet Source #
Instance details Defined in 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 TD3 Methods flattenParameters :: CriticNet -> [Parameter] # _replaceParameters :: CriticNet -> ParamStream CriticNet #
Randomizable CriticNetSpec CriticNet Source #	Critic Network Weight initialization
Instance details Defined in TD3 Methods sample :: CriticNetSpec -> IO CriticNet #
type Rep CriticNet Source #
Instance details Defined in TD3 type Rep CriticNet = D1 ('MetaData "CriticNet" "TD3" "edelwace-0.1.0.0-NlkYKOSwr4E6F4MQQAmdJ" '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 :: ActorNet Online Policy φ φ' :: ActorNet Target Policy φ' θ :: CriticNet Online Critic θ θ' :: CriticNet Target Critic θ φOptim :: Adam Policy Optimizer θOptim :: Adam Critic Optimizer

Instances

Instances details

Show Agent Source #
Instance details Defined in TD3 Methods showsPrec :: Int -> Agent -> ShowS # show :: Agent -> String # showList :: [Agent] -> ShowS #
Generic Agent Source #
Instance details Defined in TD3 Associated Types type Rep Agent :: Type -> Type # Methods from :: Agent -> Rep Agent x # to :: Rep Agent x -> Agent #
type Rep Agent Source #
Instance details Defined in TD3 type Rep Agent = D1 ('MetaData "Agent" "TD3" "edelwace-0.1.0.0-NlkYKOSwr4E6F4MQQAmdJ" 'False) (C1 ('MetaCons "Agent" 'PrefixI 'True) ((S1 ('MetaSel ('Just "\966") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 ActorNet) :: (S1 ('MetaSel ('Just "\966'") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedLazy) (Rec0 ActorNet) :: 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)))))