Torch.Initialization

Helper functions which might end up migrating to the -indef codebase

Documentation

newLinear :: forall o i. All KnownDim '[i, o] => Generator -> IO (Linear i o)

initialize a new linear layer

newConv2d :: forall o i kH kW. All KnownDim '[i, o, kH, kW, kH * kW] => Generator -> IO (Conv2d i o '(kH, kW))

initialize a new conv2d layer

xavierUniformWith_

Arguments

:: (Dimensions outs, All KnownDim '[i, o, Product outs])

=> HsReal

gain: an optional scaling factor

-> Generator

-> Tensor (i :+ (o :+ outs))

tensor: an n-dimensional `torch.Tensor` (minimum length 2)

-> IO ()

Fills the input Tensor with values according to the method described in "Understanding the difficulty of training deep feedforward neural networks" - Glorot, X. & Bengio, Y. (2010), using a uniform distribution. The resulting tensor will have values sampled from :math:`mathcal{U}(-a, a)` where .. math:: a = text{gain} times sqrt{frac{6}{text{fan_in} + text{fan_out}}} Also known as Glorot initialization. Examples: -set -XScopedTypeVariables w :: Tensor '[3, 5] <- torch.new xavierUniformWith_ w (calculate_gain Relu)

xavierUniform_

Arguments

:: (Dimensions outs, All KnownDim '[i, o, Product outs])

=> Generator

-> Tensor (i :+ (o :+ outs))

tensor: an n-dimensional `torch.Tensor` (minimum length 2)

-> IO ()

xavierUniformWith_ with default of gain = 1

xavierNormalWith_

Arguments

:: (Dimensions outs, All KnownDim '[i, o, Product outs])

=> HsReal

gain: an optional scaling factor

-> Generator

-> Tensor (i :+ (o :+ outs))

tensor: an n-dimensional `torch.Tensor` (minimum length 2)

-> IO ()

xavierNormal_

Arguments

:: (Dimensions outs, All KnownDim '[i, o, Product outs])

=> Generator

-> Tensor (i :+ (o :+ outs))

tensor: an n-dimensional `torch.Tensor` (minimum length 2)

-> IO ()

xavierNormalWith_ with default of gain = 1

data Activation

Constructors

LinearFn

Linear activation

Conv1dFn

Conv1d activation

Conv2dFn

Conv2d activation

Conv3dFn

Conv3d activation

Conv1dTFn

Conv1d transpose activation

Conv2dTFn

Conv2d transpose activation

Conv3dTFn

Conv3d transpose activation

SigmoidFn

TanhFn

ReluFn

LeakyReluFn (Maybe Double)

Instances

Eq Activation

Defined in Torch.Initialization

Methods

(==) :: Activation -> Activation -> Bool

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

Show Activation

Defined in Torch.Initialization

Methods

showsPrec :: Int -> Activation -> ShowS

show :: Activation -> String

showList :: [Activation] -> ShowS

data FanMode

Constructors

FanIn

FanOut

Instances

Eq FanMode

Defined in Torch.Initialization

Methods

(==) :: FanMode -> FanMode -> Bool

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

Ord FanMode

Defined in Torch.Initialization

Methods

compare :: FanMode -> FanMode -> Ordering

(<) :: FanMode -> FanMode -> Bool

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

(>) :: FanMode -> FanMode -> Bool

(>=) :: FanMode -> FanMode -> Bool

max :: FanMode -> FanMode -> FanMode

min :: FanMode -> FanMode -> FanMode

Show FanMode

Defined in Torch.Initialization

Methods

showsPrec :: Int -> FanMode -> ShowS

show :: FanMode -> String

showList :: [FanMode] -> ShowS

kaimingUniformWith_

Arguments

:: (Dimensions outs, All KnownDim '[i, o, Product outs])

=> Activation

-> FanMode

-> Generator

-> Tensor (i :+ (o :+ outs))

tensor: an n-dimensional `torch.Tensor` (minimum length 2)

-> IO ()

Fills the input Tensor with values according to the method described in "Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification" - He, K. et al. (2015), using a uniform distribution. The resulting tensor will have values sampled from :math:`mathcal{U}(-text{bound}, text{bound})` where .. math:: text{bound} = sqrt{frac{6}{(1 + a^2) times text{fan_in}}} Also known as He initialization. Args: tensor: an n-dimensional `torch.Tensor` a: the negative slope of the rectifier used after this layer (0 for ReLU by default) mode: either fan_in (default) or fan_out. Choosing fan_in preserves the magnitude of the variance of the weights in the forward pass. Choosing fan_out preserves the magnitudes in the backwards pass. nonlinearity: the non-linear function (`nn.functional` name), recommended to use only with relu or leaky_relu (default). Examples: >>> w = torch.empty(3, 5) >>> nn.init.kaiming_uniform_(w, mode=fan_in, nonlinearity=relu)

kaimingUniform_

Arguments

:: (Dimensions outs, All KnownDim '[i, o, Product outs])

=> Generator

-> Tensor (i :+ (o :+ outs))

tensor: an n-dimensional `torch.Tensor` (minimum length 2)

-> IO ()

kaimingNormalWith_

Arguments

:: (Dimensions outs, All KnownDim '[i, o, Product outs])

=> Activation

-> FanMode

-> Generator

-> Tensor (i :+ (o :+ outs))

tensor: an n-dimensional `torch.Tensor` (minimum length 2)

-> IO ()

Fills the input Tensor with values according to the method described in "Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification" - He, K. et al. (2015), using a normal distribution. The resulting tensor will have values sampled from :math:`mathcal{N}(0, text{std})` where .. math:: text{std} = sqrt{frac{2}{(1 + a^2) times text{fan_in}}} Also known as He initialization. Args: tensor: an n-dimensional `torch.Tensor` a: the negative slope of the rectifier used after this layer (0 for ReLU by default) mode: either fan_in (default) or fan_out. Choosing fan_in preserves the magnitude of the variance of the weights in the forward pass. Choosing fan_out preserves the magnitudes in the backwards pass. nonlinearity: the non-linear function (`nn.functional` name), recommended to use only with relu or leaky_relu (default). Examples: >>> w = torch.empty(3, 5) >>> nn.init.kaiming_normal_(w, mode=fan_out, nonlinearity=relu)

kaimingNormal_

Arguments

:: (Dimensions outs, All KnownDim '[i, o, Product outs])

=> Generator

-> Tensor (i :+ (o :+ outs))

tensor: an n-dimensional `torch.Tensor` (minimum length 2)

-> IO ()