Safe Haskell	Safe-Inferred
Language	Haskell2010

Torch.GraduallyTyped.Optim

Synopsis

data AdamOptions = AdamOptions {
- learningRate :: Double
- beta1 :: Double
- beta2 :: Double
- epsilon :: Double
- weightDecay :: Double
- amsgrad :: Bool
}
defaultAdamOptions :: AdamOptions
data Optimizer model where
- UnsafeOptimizer :: forall model. {..} -> Optimizer model
getStateDict :: forall model. Optimizer model -> IO StateDict
getModel :: forall model. HasStateDict model => ModelSpec model -> Optimizer model -> IO model
mkAdam :: forall model. HasStateDict model => AdamOptions -> model -> IO (Optimizer model)
stepWithGenerator :: forall model generatorDevice lossGradient lossLayout lossDataType lossDevice lossShape generatorOutputDevice. (HasStateDict model, SGetGeneratorDevice generatorDevice, SGetGeneratorDevice generatorOutputDevice, Catch (lossShape <+> 'Shape '[]), Catch (lossGradient <+> 'Gradient 'WithGradient)) => Optimizer model -> ModelSpec model -> (model -> Generator generatorDevice -> IO (Tensor lossGradient lossLayout lossDataType lossDevice lossShape, Generator generatorOutputDevice)) -> Generator generatorDevice -> IO (Tensor lossGradient lossLayout lossDataType lossDevice lossShape, Generator generatorOutputDevice)

Documentation

Options for the Adam optimizer.

Constructors

AdamOptions
Fields learningRate :: Double learning rate beta1 :: Double beta1 beta2 :: Double beta2 epsilon :: Double epsilon weightDecay :: Double weight decay amsgrad :: Bool use amsgrad

Default Adam options.

data Optimizer model where Source #

Optimizer data type.

Constructors

UnsafeOptimizer
Fields :: forall model. { optimizerStateDictKeys :: [StateDictKey] , optimizerPtr :: ForeignPtr Optimizer } -> Optimizer model

Get the model state dictionary from an optimizer.

getModel :: forall model. HasStateDict model => ModelSpec model -> Optimizer model -> IO model Source #

Extract a model from an optimizer.

Arguments

:: forall model. HasStateDict model
=> AdamOptions	Adam options
-> model	initial model
-> IO (Optimizer model)	Adam optimizer

Create a new Adam optimizer from a model.

Arguments

:: forall model generatorDevice lossGradient lossLayout lossDataType lossDevice lossShape generatorOutputDevice. (HasStateDict model, SGetGeneratorDevice generatorDevice, SGetGeneratorDevice generatorOutputDevice, Catch (lossShape <+> 'Shape '[]), Catch (lossGradient <+> 'Gradient 'WithGradient))
=> Optimizer model	optimizer for the model
-> ModelSpec model	model specification
-> (model -> Generator generatorDevice -> IO (Tensor lossGradient lossLayout lossDataType lossDevice lossShape, Generator generatorOutputDevice))	loss function to minimize
-> Generator generatorDevice	random generator
-> IO (Tensor lossGradient lossLayout lossDataType lossDevice lossShape, Generator generatorOutputDevice)	loss and updated generator

Perform one step of optimization.