NN.Architectures.LossFunctions

Definitions

Loss : AddCont -> AddCont -> Type

  A loss is a parametric map whose parameter is the label
  It is not a `Model` because there is no concept of initialisation

Totality: total
Visibility: public export

Label : Loss y l -> Type

  The "parameter" object of the loss function is the type of supervised
  learning labels

Totality: total
Visibility: public export

pairLossFunctions : {auto {conArg:898} : Num l} -> Loss y (Const l) -> Loss z (Const l) -> Loss (y >*< z) (Const l)

  Run two losses in parallel, add their results

Totality: total
Visibility: public export

chosenBranchLoss : ((i : c .Pos (shapeExt branches)) -> Loss (index branches i) l) -> Loss (Section (index branches)) l

  The loss of a section against a labelled branch
  We always choose the branch the label chooses

Totality: total
Visibility: public export

matchedBranchLoss : IsDecidable c => ((i : c .Pos (shapeExt branches)) -> Loss (index branches i) l) -> Loss (Coproduct branches) (Maybe l)

  Same as above, except we take an already chosen branch, and fail if 
  label disagrees with it

Totality: total
Visibility: public export

madeChoice : {auto {conArg:1110} : IsNaperian ax} -> DecEq (Log ax) => AddContDPair f =%+> (Simplex ax >*< AddContDPair f)

  Given a made choice, pair with with a dirac delta distribution

Totality: total
Visibility: public export

SquaredError : {auto {conArg:1170} : Num a} -> Neg a => Loss (Const a) (Const a)

Totality: total
Visibility: public export

MeanSquaredError : IsCubical n => {auto {conArg:1214} : TensorMonoid (n .cont)} -> {auto {conArg:1220} : Num a} -> Neg a => Fractional a => Cast Nat a => Loss (Const (Tensor [n] a)) (Const (Tensor [] a))

Totality: total
Visibility: public export

softargmaxCrossEntropyLogits : IsNaperian a => IsFoldable (a .cont) => (Simplex a >*< Simplex a) =%+> Const (Tensor [] Double)

  The payoff object is the rank-0 tensor, not `Double`
  Defined for Naperian axes, where the prediction and the label have one shape
  TODO the prediction and the label can be distributions of different shapes!

Totality: total
Visibility: public export

SoftargmaxCrossEntropyLogits : IsNaperian a => IsFoldable (a .cont) => Loss (Simplex a) (Const (Tensor [] Double))

Totality: total
Visibility: public export