Safe Haskell	None
Language	Haskell2010

RPB

Description

General Replay Buffer Types and TypeClasses

Synopsis

class Functor b => ReplayBuffer b where
- size :: b Tensor -> Int
- push :: Int -> b Tensor -> b Tensor -> b Tensor
- lookUp :: [Int] -> b Tensor -> b Tensor
- sampleIO :: Int -> b Tensor -> IO (b Tensor)
- asTuple :: b Tensor -> (Tensor, Tensor, Tensor, Tensor, Tensor)
- collectExperience :: Agent a => Params -> CircusUrl -> Tracker -> Int -> a -> IO (b Tensor)
randomBatches :: ReplayBuffer b => Int -> Int -> b Tensor -> IO [Transition]
data Buffer a = Buffer {
- states :: !a
- actions :: !a
- rewards :: !a
- states' :: !a
- dones :: !a
}
empty :: Buffer Tensor
size' :: Buffer Tensor -> Int
drop :: Int -> Buffer Tensor -> Buffer Tensor
push' :: Int -> Buffer Tensor -> Buffer Tensor -> Buffer Tensor
lookUp' :: [Int] -> Buffer Tensor -> Buffer Tensor
sampleIO' :: Int -> Buffer Tensor -> IO (Buffer Tensor)
asTuple' :: Buffer Tensor -> Transition
collectStep :: Agent a => Params -> CircusUrl -> Tracker -> Int -> Int -> a -> Tensor -> Buffer Tensor -> IO (Buffer Tensor)
collectExperience' :: Agent a => Params -> CircusUrl -> Tracker -> Int -> a -> IO (Buffer Tensor)

Documentation

class Functor b => ReplayBuffer b where Source #

Replay Buffer Interface

Methods

size :: b Tensor -> Int Source #

Return size of current buffer

push :: Int -> b Tensor -> b Tensor -> b Tensor Source #

Push one buffer into another

lookUp :: [Int] -> b Tensor -> b Tensor Source #

Look Up given list if indices

sampleIO :: Int -> b Tensor -> IO (b Tensor) Source #

Take n Random Samples

asTuple :: b Tensor -> (Tensor, Tensor, Tensor, Tensor, Tensor) Source #

Return the Tuple: (s, a, r, s', d) for training

collectExperience :: Agent a => Params -> CircusUrl -> Tracker -> Int -> a -> IO (b Tensor) Source #

Collect Experiences in Buffer

Instances

Instances details

ReplayBuffer Buffer Source #	Vanilla Replay Buffer implements `ReplayBuffer`
Instance details Defined in RPB Methods size :: Buffer Tensor -> Int Source # push :: Int -> Buffer Tensor -> Buffer Tensor -> Buffer Tensor Source # lookUp :: [Int] -> Buffer Tensor -> Buffer Tensor Source # sampleIO :: Int -> Buffer Tensor -> IO (Buffer Tensor) Source # asTuple :: Buffer Tensor -> (Tensor, Tensor, Tensor, Tensor, Tensor) Source # collectExperience :: Agent a => Params -> CircusUrl -> Tracker -> Int -> a -> IO (Buffer Tensor) Source #
ReplayBuffer Buffer Source #	Hindsight Experience Replay Buffer implements `ReplayBuffer`
Instance details Defined in RPB.HER Methods size :: Buffer Tensor -> Int Source # push :: Int -> Buffer Tensor -> Buffer Tensor -> Buffer Tensor Source # lookUp :: [Int] -> Buffer Tensor -> Buffer Tensor Source # sampleIO :: Int -> Buffer Tensor -> IO (Buffer Tensor) Source # asTuple :: Buffer Tensor -> (Tensor, Tensor, Tensor, Tensor, Tensor) Source # collectExperience :: Agent a => Params -> CircusUrl -> Tracker -> Int -> a -> IO (Buffer Tensor) Source #

randomBatches :: ReplayBuffer b => Int -> Int -> b Tensor -> IO [Transition] Source #

Generate a list of uniformly sampled minibatches

data Buffer a Source #

Vanilla Replay Buffer

Constructors

Buffer
Fields states :: !a States actions :: !a Actions rewards :: !a Rewards states' :: !a Next States dones :: !a Terminal Mask

Instances

Instances details

Functor Buffer Source #	Vanilla ReplayBuffer implements `functor`
Instance details Defined in RPB Methods fmap :: (a -> b) -> Buffer a -> Buffer b # (<$) :: a -> Buffer b -> Buffer a #
Applicative Buffer Source #	This is badly defined and only so it can use `liftA2`.
Instance details Defined in RPB Methods pure :: a -> Buffer a # (<>) :: Buffer (a -> b) -> Buffer a -> Buffer b # liftA2 :: (a -> b -> c) -> Buffer a -> Buffer b -> Buffer c # (>) :: Buffer a -> Buffer b -> Buffer b # (<*) :: Buffer a -> Buffer b -> Buffer a #
ReplayBuffer Buffer Source #	Vanilla Replay Buffer implements `ReplayBuffer`
Instance details Defined in RPB Methods size :: Buffer Tensor -> Int Source # push :: Int -> Buffer Tensor -> Buffer Tensor -> Buffer Tensor Source # lookUp :: [Int] -> Buffer Tensor -> Buffer Tensor Source # sampleIO :: Int -> Buffer Tensor -> IO (Buffer Tensor) Source # asTuple :: Buffer Tensor -> (Tensor, Tensor, Tensor, Tensor, Tensor) Source # collectExperience :: Agent a => Params -> CircusUrl -> Tracker -> Int -> a -> IO (Buffer Tensor) Source #
Eq a => Eq (Buffer a) Source #
Instance details Defined in RPB Methods (==) :: Buffer a -> Buffer a -> Bool # (/=) :: Buffer a -> Buffer a -> Bool #
Show a => Show (Buffer a) Source #
Instance details Defined in RPB Methods showsPrec :: Int -> Buffer a -> ShowS # show :: Buffer a -> String # showList :: [Buffer a] -> ShowS #

empty :: Buffer Tensor Source #

Create a new, empty Buffer on the CPU

size' :: Buffer Tensor -> Int Source #

How many Trajectories are currently stored in memory

drop :: Int -> Buffer Tensor -> Buffer Tensor Source #

Drop number of entries from the beginning of the Buffer

push' :: Int -> Buffer Tensor -> Buffer Tensor -> Buffer Tensor Source #

Push one buffer into another one

lookUp' :: [Int] -> Buffer Tensor -> Buffer Tensor Source #

Get the given indices from Buffer

sampleIO' :: Int -> Buffer Tensor -> IO (Buffer Tensor) Source #

Take n random samples from Buffer

asTuple' :: Buffer Tensor -> Transition Source #

Return (State, Action, Reward, Next State, Done) Tuple

collectStep :: Agent a => Params -> CircusUrl -> Tracker -> Int -> Int -> a -> Tensor -> Buffer Tensor -> IO (Buffer Tensor) Source #

Evaluate Policy for T steps and return experience Buffer

collectExperience' :: Agent a => Params -> CircusUrl -> Tracker -> Int -> a -> IO (Buffer Tensor) Source #

Collect experience for a given number of steps