Safe Haskell	Safe-Inferred
Language	Haskell2010

Torch.GraduallyTyped.NN.Functional.NonLinearActivation

Synopsis

type SoftMaxErrorMessage (by :: By Symbol Nat) (dims :: [Dim (Name Symbol) (Size Nat)]) = "Cannot apply softmax on the dimension matching" % ("" % (((" '" <> by) <> "'") % ("" % ("in the shape" % ("" % (((" '" <> dims) <> "'.") % ""))))))
type family SoftmaxCheckF (by :: By Symbol Nat) (dims :: [Dim (Name Symbol) (Size Nat)]) (result :: Maybe (Dim (Name Symbol) (Size Nat))) :: [Dim (Name Symbol) (Size Nat)] where ...
type family SoftmaxF (selectDim :: SelectDim (By Symbol Nat)) (shape :: Shape [Dim (Name Symbol) (Size Nat)]) :: Shape [Dim (Name Symbol) (Size Nat)] where ...
softmax :: forall selectDim gradient layout device dataType shape shape' m. (MonadThrow m, shape' ~ SoftmaxF selectDim shape, Catch shape') => SSelectDim selectDim -> Tensor gradient layout device dataType shape -> m (Tensor gradient layout device dataType shape')
logSoftmax :: forall selectDim gradient layout device dataType shape shape' m. (MonadThrow m, shape' ~ SoftmaxF selectDim shape, Catch shape') => SSelectDim selectDim -> Tensor gradient layout device dataType shape -> m (Tensor gradient layout device dataType shape')

Documentation

>>> import Torch.GraduallyTyped.Prelude.List (SList (..))
>>> import Torch.GraduallyTyped

type SoftMaxErrorMessage (by :: By Symbol Nat) (dims :: [Dim (Name Symbol) (Size Nat)]) = "Cannot apply softmax on the dimension matching" % ("" % (((" '" <> by) <> "'") % ("" % ("in the shape" % ("" % (((" '" <> dims) <> "'.") % "")))))) Source #

type family SoftmaxCheckF (by :: By Symbol Nat) (dims :: [Dim (Name Symbol) (Size Nat)]) (result :: Maybe (Dim (Name Symbol) (Size Nat))) :: [Dim (Name Symbol) (Size Nat)] where ... Source #

Equations

SoftmaxCheckF by dims 'Nothing = TypeError (SoftMaxErrorMessage by dims)
SoftmaxCheckF _ dims ('Just _) = dims

type family SoftmaxF (selectDim :: SelectDim (By Symbol Nat)) (shape :: Shape [Dim (Name Symbol) (Size Nat)]) :: Shape [Dim (Name Symbol) (Size Nat)] where ... Source #

Equations

SoftmaxF 'UncheckedSelectDim _ = 'UncheckedShape
SoftmaxF _ 'UncheckedShape = 'UncheckedShape
SoftmaxF ('SelectDim by) ('Shape dims) = 'Shape (SoftmaxCheckF by dims (GetDimImplF by dims))

softmax :: forall selectDim gradient layout device dataType shape shape' m. (MonadThrow m, shape' ~ SoftmaxF selectDim shape, Catch shape') => SSelectDim selectDim -> Tensor gradient layout device dataType shape -> m (Tensor gradient layout device dataType shape') Source #

Applies the softmax function that is defined as:

\[ \mathrm{Softmax}(\mathrm{input}_{i}) = \frac{\exp\left(\mathrm{input}_{i}\right)}{\sum_j \exp\left(\mathrm{input}_{j}\right)} \]

Softmax is applied to all slices along selectDim, and will re-scale them so that the elements lie in the range \([0, 1]\) and sum to \(1\):

>>> g <- sMkGenerator (SDevice SCPU) 0
>>> (input, _) <- sRandn (TensorSpec (SGradient SWithGradient) (SLayout SDense) (SDevice SCPU) (SDataType SFloat) (SShape $ SName @"batch" :&: SSize @32 :|: SName @"feature" :&: SSize @8 :|: SNil)) g
>>> result <- softmax (SSelectDim (SByName @"feature")) input
>>> :type result
result
  :: Tensor
       ('Gradient 'WithGradient)
       ('Layout 'Dense)
       ('Device 'CPU)
       ('DataType 'Float)
       ('Shape
          '[ 'Dim ('Name "batch") ('Size 32),
             'Dim ('Name "feature") ('Size 8)])

logSoftmax :: forall selectDim gradient layout device dataType shape shape' m. (MonadThrow m, shape' ~ SoftmaxF selectDim shape, Catch shape') => SSelectDim selectDim -> Tensor gradient layout device dataType shape -> m (Tensor gradient layout device dataType shape') Source #

Applies the softmax function that is defined as:

\[ \mathrm{Softmax}(\mathrm{input}_{i}) = \frac{\exp\left(\mathrm{input}_{i}\right)}{\sum_j \exp\left(\mathrm{input}_{j}\right)} \]

Softmax is applied to all slices along selectDim, and will re-scale them so that the elements lie in the range \([0, 1]\) and sum to \(1\):

>>> g <- sMkGenerator (SDevice SCPU) 0
>>> (input, _) <- sRandn (TensorSpec (SGradient SWithGradient) (SLayout SDense) (SDevice SCPU) (SDataType SFloat) (SShape $ SName @"batch" :&: SSize @32 :|: SName @"feature" :&: SSize @8 :|: SNil)) g
>>> result <- softmax (SSelectDim (SByName @"feature")) input
>>> :type result
result
  :: Tensor
       ('Gradient 'WithGradient)
       ('Layout 'Dense)
       ('Device 'CPU)
       ('DataType 'Float)
       ('Shape
          '[ 'Dim ('Name "batch") ('Size 32),
             'Dim ('Name "feature") ('Size 8)])