Data.Autodiff.Ops

Reexports

Definitions

mulModel : {auto {conArg:10508} : Neg t} -> Random t => Const t -\-> Const t

Totality: total
Visibility: public export

addModel : {auto {conArg:10567} : Neg t} -> Random t => Const t -\-> Const t

Totality: total
Visibility: public export

scalarAffine : {auto {conArg:10626} : Neg t} -> Random t => Materialise t => Const t -\-> Const t

Totality: total
Visibility: public export

parallelTensor : {auto {conArg:10685} : Num a} -> {auto {conArg:10688} : Num b} -> {auto {conArg:10693} : AllC TensorMonoid shape} -> AllC IsConcrete shape => Const a =%+> Const b -> Const (Tensor shape a) =%+> Const (Tensor shape b)

  Apply a scalar lens elementwise across a tensor

Totality: total
Visibility: public export

parallelTensorModel : {auto {conArg:10854} : Num a} -> {auto {conArg:10857} : Num b} -> {auto {conArg:10862} : AllC TensorMonoid shape} -> AllC IsConcrete shape => Traversable (Tensor shape) => Const a -\-> Const b -> Const (Tensor shape a) -\-> Const (Tensor shape b)

  `Tensor shape` applied to a model: one copy per entry, each with its own parameter

Totality: total
Visibility: public export

copyN : {auto {conArg:11147} : Num a} -> Const a =%+> Const (Tensor [(axisName ~~> n)] a)

Totality: total
Visibility: public export

sameFromTensorN : {auto {conArg:11264} : Num a} -> {auto {conArg:11267} : Num b} -> Traversable (Tensor [(axisName ~~> n)]) => Const a -\-> Const b -> Const a -\-> Const (Tensor [(axisName ~~> n)] b)

Totality: total
Visibility: public export

sumAxis : IsCubical n => {auto {conArg:11410} : Num a} -> {auto {conArg:11413} : TensorMonoid (n .cont)} -> Const (Tensor [n] a) =%+> Const (Tensor [] a)

  Dual to `copyN`

Totality: total
Visibility: public export

divBy : {auto {conArg:11537} : Num a} -> Fractional a => a -> Const (Tensor [] a) =%+> Const (Tensor [] a)

  Divide by a constant, entrywise in both directions

Totality: total
Visibility: public export

meanSquaredDifference : IsCubical n => {auto {conArg:11620} : TensorMonoid (n .cont)} -> {auto {conArg:11626} : Num a} -> Neg a => Fractional a => Cast Nat a => (Const (Tensor [n] a) >*< Const (Tensor [n] a)) =%+> Const (Tensor [] a)

Totality: total
Visibility: public export

leakyReLU : {auto {conArg:11794} : Num a} -> Ord a => a -> Const a =%+> Const a

  Recovers `ReLU` when `alpha=0`
  Cannot be written as a composition of scaling and `ReLU`

Totality: total
Visibility: public export

leakyReLUModel : {auto {conArg:11912} : Num a} -> Ord a => a -> {auto {conArg:11922} : AllC TensorMonoid shape} -> AllC IsConcrete shape => Const (Tensor shape a) -\-> Const (Tensor shape a)

Totality: total
Visibility: public export

reluModel : {auto {conArg:12001} : Num a} -> Ord a => {auto {conArg:12009} : AllC TensorMonoid shape} -> AllC IsConcrete shape => Const (Tensor shape a) -\-> Const (Tensor shape a)

Totality: total
Visibility: public export

fromLogits : Const (Tensor [(name ~~> n)] Double) =%+> Simplex name n

  Interpret a vector as logits of a distribution. The backward pass is
  identity: gradients are computed in the sme way

Totality: total
Visibility: public export

fromLogitsModel : Const (Tensor [(name ~~> n)] Double) -\-> Simplex name n

Totality: total
Visibility: public export