Objective Function

In oneDAL, the objective function represents an interface of objective functions K(θ)=F(θ)+M(θ), where F(θ) is a smooth and M(θ) is a non-smooth functions, that accepts input argument θRp and returns:

  • The value of objective function, y=K(θ)

  • The value of M(θ), yns=M(θ)

  • The gradient of F(θ):

    g(θ)=F(θ)={Fθ1,,Fθp}
  • The Hessian of F(θ):

    H==2F(θ)=2Fi=[Fiθ1θ1Fiθ1θpFipθ1Fiθpθp]
  • The objective function specific projection of proximal operator (see [MSE, Log-Loss, Cross-Entropy] for details):

    proxηM(x)=argminuRp(M(u)+12η|ux|22)xRp
  • The objective function specific Lipschitz constant, constantOfLipschitz||F(θ).

Note

On GPU, only Logistic Loss and Cross-entropy Loss are supported, Mean Squared Error Algorithm is not supported.