Numeric.SGD.Momentum

Provides the momentum function which implements stochastic gradient descent with momentum, following:

• http://ruder.io/optimizing-gradient-descent/index.html#momentum

Documentation

data Config

Momentum configuration

Constructors

Config

Fields

• alpha0 :: Double

  Initial step size, used to scale the gradient

• tau :: Double

  The step size after k * tau iterations = alpha0 / (k + 1)

• gamma :: Double

  Momentum term

Instances

Eq Config

Defined in Numeric.SGD.Momentum

Methods

(==) :: Config -> Config -> Bool

(/=) :: Config -> Config -> Bool

Ord Config

Defined in Numeric.SGD.Momentum

Methods

compare :: Config -> Config -> Ordering

(<) :: Config -> Config -> Bool

(<=) :: Config -> Config -> Bool

(>) :: Config -> Config -> Bool

(>=) :: Config -> Config -> Bool

max :: Config -> Config -> Config

min :: Config -> Config -> Config

Show Config

Defined in Numeric.SGD.Momentum

Methods

showsPrec :: Int -> Config -> ShowS

show :: Config -> String

showList :: [Config] -> ShowS

Generic Config

Defined in Numeric.SGD.Momentum

Associated Types

type Rep Config :: Type -> Type

Methods

from :: Config -> Rep Config x

to :: Rep Config x -> Config

Default Config

Defined in Numeric.SGD.Momentum

Methods

def :: Config

type Rep Config

Defined in Numeric.SGD.Momentum

type Rep Config = D1 (MetaData "Config" "Numeric.SGD.Momentum" "sgd-0.8.0.3-2sUF012Uk8gIqgMlYiSV0v" False) (C1 (MetaCons "Config" PrefixI True) (S1 (MetaSel (Just "alpha0") NoSourceUnpackedness NoSourceStrictness DecidedLazy) (Rec0 Double) :*: (S1 (MetaSel (Just "tau") NoSourceUnpackedness NoSourceStrictness DecidedLazy) (Rec0 Double) :*: S1 (MetaSel (Just "gamma") NoSourceUnpackedness NoSourceStrictness DecidedLazy) (Rec0 Double))))

scaleTau :: Double -> Config -> Config

Scale the tau parameter. Useful e.g. to account for the size of the training dataset.

momentum

Arguments

:: (Monad m, ParamSet p)

=> Config

Momentum configuration

-> (e -> p -> p)

Gradient on a training element

-> SGD m e p

Stochastic gradient descent with momentum. See Numeric.SGD.Momentum for more information.