Safe Haskell	None
Language	Haskell2010

Torch.Typed.NN.Linear

Synopsis

data LinearSpec (inputFeatures :: Nat) (outputFeatures :: Nat) (dtype :: DType) (device :: (DeviceType, Nat)) = LinearSpec
data Linear (inputFeatures :: Nat) (outputFeatures :: Nat) (dtype :: DType) (device :: (DeviceType, Nat)) where
- Linear :: forall (inputFeatures :: Nat) (outputFeatures :: Nat) (dtype :: DType) (device :: (DeviceType, Nat)). {..} -> Linear inputFeatures outputFeatures dtype device
linearForward :: forall {inputFeatures :: Nat} {outputFeatures :: Nat} {w1 :: DType} {w2 :: (DeviceType, Nat)} {shape :: [Nat]}. Linear inputFeatures outputFeatures w1 w2 -> Tensor w2 w1 shape -> Tensor w2 w1 (CheckBroadcast (CheckMatMul shape '[inputFeatures, outputFeatures] (ComputeMatMul (ReverseImpl shape ('[] :: [Nat])) '[outputFeatures, inputFeatures])) (CheckMatMul shape '[inputFeatures, outputFeatures] (ComputeMatMul (ReverseImpl shape ('[] :: [Nat])) '[outputFeatures, inputFeatures])) (ComputeBroadcast (ReverseImpl (CheckMatMul shape '[inputFeatures, outputFeatures] (ComputeMatMul (ReverseImpl shape ('[] :: [Nat])) '[outputFeatures, inputFeatures])) ('[] :: [Nat])) (ReverseImpl (CheckMatMul shape '[inputFeatures, outputFeatures] (ComputeMatMul (ReverseImpl shape ('[] :: [Nat])) '[outputFeatures, inputFeatures])) ('[] :: [Nat]))))

Documentation

data LinearSpec (inputFeatures :: Nat) (outputFeatures :: Nat) (dtype :: DType) (device :: (DeviceType, Nat)) Source #

Constructors

LinearSpec

Instances

Instances details

Show (LinearSpec inputFeatures outputFeatures dtype device) Source #
Instance details Defined in Torch.Typed.NN.Linear Methods showsPrec :: Int -> LinearSpec inputFeatures outputFeatures dtype device -> ShowS # show :: LinearSpec inputFeatures outputFeatures dtype device -> String # showList :: [LinearSpec inputFeatures outputFeatures dtype device] -> ShowS #
Eq (LinearSpec inputFeatures outputFeatures dtype device) Source #
Instance details Defined in Torch.Typed.NN.Linear Methods (==) :: LinearSpec inputFeatures outputFeatures dtype device -> LinearSpec inputFeatures outputFeatures dtype device -> Bool # (/=) :: LinearSpec inputFeatures outputFeatures dtype device -> LinearSpec inputFeatures outputFeatures dtype device -> Bool #
(KnownNat inputFeatures, KnownNat outputFeatures, KnownDType dtype, KnownDevice device, RandDTypeIsValid device dtype) => Randomizable (LinearSpec inputFeatures outputFeatures dtype device) (Linear inputFeatures outputFeatures dtype device) Source #
Instance details Defined in Torch.Typed.NN.Linear Methods sample :: LinearSpec inputFeatures outputFeatures dtype device -> IO (Linear inputFeatures outputFeatures dtype device) Source #

data Linear (inputFeatures :: Nat) (outputFeatures :: Nat) (dtype :: DType) (device :: (DeviceType, Nat)) where Source #

Constructors

Linear
Fields :: forall (inputFeatures :: Nat) (outputFeatures :: Nat) (dtype :: DType) (device :: (DeviceType, Nat)). { weight :: Parameter device dtype '[outputFeatures, inputFeatures] , bias :: Parameter device dtype '[outputFeatures] } -> Linear inputFeatures outputFeatures dtype device

Instances

Instances details

Generic (Linear inputFeatures outputFeatures dtype device) Source #

Instance details

Defined in Torch.Typed.NN.Linear

Associated Types

type Rep (Linear inputFeatures outputFeatures dtype device)

Instance details

Defined in Torch.Typed.NN.Linear

type Rep (Linear inputFeatures outputFeatures dtype device) = D1 ('MetaData "Linear" "Torch.Typed.NN.Linear" "hasktorch-0.2.1.2-inplace" 'False) (C1 ('MetaCons "Linear" 'PrefixI 'True) (S1 ('MetaSel ('Just "weight") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedStrict) (Rec0 (Parameter device dtype '[outputFeatures, inputFeatures])) :*: S1 ('MetaSel ('Just "bias") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedStrict) (Rec0 (Parameter device dtype '[outputFeatures]))))

Methods

from :: Linear inputFeatures outputFeatures dtype device -> Rep (Linear inputFeatures outputFeatures dtype device) x #

to :: Rep (Linear inputFeatures outputFeatures dtype device) x -> Linear inputFeatures outputFeatures dtype device #

Show (Linear inputFeatures outputFeatures dtype device) Source #

Instance details

Defined in Torch.Typed.NN.Linear

Methods

showsPrec :: Int -> Linear inputFeatures outputFeatures dtype device -> ShowS #

show :: Linear inputFeatures outputFeatures dtype device -> String #

showList :: [Linear inputFeatures outputFeatures dtype device] -> ShowS #

Parameterized (Linear inputFeatures outputFeatures dtype device) Source #

Instance details

Defined in Torch.Typed.NN.Linear

Associated Types

type Parameters (Linear inputFeatures outputFeatures dtype device)
Instance details Defined in Torch.Typed.NN.Linear type Parameters (Linear inputFeatures outputFeatures dtype device) = GParameters (Rep (Linear inputFeatures outputFeatures dtype device))

Methods

flattenParameters :: Linear inputFeatures outputFeatures dtype device -> HList (Parameters (Linear inputFeatures outputFeatures dtype device)) Source #

replaceParameters :: Linear inputFeatures outputFeatures dtype device -> HList (Parameters (Linear inputFeatures outputFeatures dtype device)) -> Linear inputFeatures outputFeatures dtype device Source #

(shape'' ~ MatMul shape '[inputFeatures, outputFeatures], shape' ~ Broadcast shape'' shape'') => HasForward (Linear inputFeatures outputFeatures dtype device) (Tensor device dtype shape) (Tensor device dtype shape') Source #

Instance details

Defined in Torch.Typed.NN.Linear

Methods

forward :: Linear inputFeatures outputFeatures dtype device -> Tensor device dtype shape -> Tensor device dtype shape' Source #

forwardStoch :: Linear inputFeatures outputFeatures dtype device -> Tensor device dtype shape -> IO (Tensor device dtype shape') Source #

(KnownNat inputFeatures, KnownNat outputFeatures, KnownDType dtype, KnownDevice device, RandDTypeIsValid device dtype) => Randomizable (LinearSpec inputFeatures outputFeatures dtype device) (Linear inputFeatures outputFeatures dtype device) Source #

Instance details

Defined in Torch.Typed.NN.Linear

Methods

sample :: LinearSpec inputFeatures outputFeatures dtype device -> IO (Linear inputFeatures outputFeatures dtype device) Source #

type Rep (Linear inputFeatures outputFeatures dtype device) Source #

Instance details

Defined in Torch.Typed.NN.Linear

type Rep (Linear inputFeatures outputFeatures dtype device) = D1 ('MetaData "Linear" "Torch.Typed.NN.Linear" "hasktorch-0.2.1.2-inplace" 'False) (C1 ('MetaCons "Linear" 'PrefixI 'True) (S1 ('MetaSel ('Just "weight") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedStrict) (Rec0 (Parameter device dtype '[outputFeatures, inputFeatures])) :*: S1 ('MetaSel ('Just "bias") 'NoSourceUnpackedness 'NoSourceStrictness 'DecidedStrict) (Rec0 (Parameter device dtype '[outputFeatures]))))

type Parameters (Linear inputFeatures outputFeatures dtype device) Source #

Instance details

Defined in Torch.Typed.NN.Linear

type Parameters (Linear inputFeatures outputFeatures dtype device) = GParameters (Rep (Linear inputFeatures outputFeatures dtype device))

linearForward :: forall {inputFeatures :: Nat} {outputFeatures :: Nat} {w1 :: DType} {w2 :: (DeviceType, Nat)} {shape :: [Nat]}. Linear inputFeatures outputFeatures w1 w2 -> Tensor w2 w1 shape -> Tensor w2 w1 (CheckBroadcast (CheckMatMul shape '[inputFeatures, outputFeatures] (ComputeMatMul (ReverseImpl shape ('[] :: [Nat])) '[outputFeatures, inputFeatures])) (CheckMatMul shape '[inputFeatures, outputFeatures] (ComputeMatMul (ReverseImpl shape ('[] :: [Nat])) '[outputFeatures, inputFeatures])) (ComputeBroadcast (ReverseImpl (CheckMatMul shape '[inputFeatures, outputFeatures] (ComputeMatMul (ReverseImpl shape ('[] :: [Nat])) '[outputFeatures, inputFeatures])) ('[] :: [Nat])) (ReverseImpl (CheckMatMul shape '[inputFeatures, outputFeatures] (ComputeMatMul (ReverseImpl shape ('[] :: [Nat])) '[outputFeatures, inputFeatures])) ('[] :: [Nat])))) Source #

linear The constraints on this one are _very_ involved, so the partial signatures make the code significantly cleaner.