Safe Haskell	Safe-Inferred
Language	Haskell2010

DataFrame.LinearSolver.Loss

Description

Smooth losses for the proximal-gradient engine. Each carries its derivative ∂ℓ/∂z at z = w·x + b and a global bound on the curvature ∂²ℓ/∂z² (used for the FISTA step size).

Synopsis

data SmoothLoss = SmoothLoss {
- slName :: !Text
- slGradZ :: Double -> Double -> Double
- slCurvBound :: !Double
}
sigmoid :: Double -> Double
logisticLoss :: SmoothLoss
squaredLoss :: SmoothLoss
sqHingeLoss :: SmoothLoss

Documentation

data SmoothLoss Source #

A convex, C¹ per-sample loss ℓ(y, z). slGradZ is ∂ℓ/∂z; slCurvBound bounds ∂²ℓ/∂z² over all (y, z).

Constructors

SmoothLoss
Fields slName :: !Text slGradZ :: Double -> Double -> Double slCurvBound :: !Double

sigmoid :: Double -> Double Source #

Numerically stable logistic sigmoid.

logisticLoss :: SmoothLoss Source #

Binary logistic loss for labels in {-1,+1}: ℓ = log(1 + exp(-y z)).

squaredLoss :: SmoothLoss Source #

Squared error for regression: ℓ = ½ (z - y)².

sqHingeLoss :: SmoothLoss Source #

Squared hinge for classification (LinearSVC default), labels {-1,+1}.