Library of functions

In the library folder a set of commonly used functions and linear operators is provided that can be used to formulate problems (see also How to use ForBES). Each of these functions is provided as a constructor taking as input all the parameters that define the function, and returning an object: this will contain all the information needed to compute the value, the gradient or the proximal mapping of the correspondent function or its conjugate.

You can access the help file of each of these function directly from MATLAB with:

help <function-name>


Name Description
distBall_l2 Distance from a Euclidean ball
distBox Distance from a box
distPos Distance from the nonnegative orthant
distNeg Distance from the nonpositive orthant
dist2Ball_l2 Squared distance from a Euclidean ball
dist2Box Squared distance from a box
dist2Pos Squared distance from the nonnegative orthant
dist2Neg Squared distance from the nonpositive orthant

Indicator functions

Name Description
indZero Indicator of the singleton set containing the origin
indBall_l0 Indicator of an L0 pseudo-norm ball
indBall_l2 Indicator of a Euclidean ball
indBox Indicator of a box
indPos Inticator of the nonnegative orthant
indNeg Indicator of the nonpositive orthant
indRankBall Indicator of the set of matrices with given rank
indBin Indicator function of a binary set

Norms and regularization functions

Name Description
l0Norm L0 pseudo-norm
l1Norm L1 norm (sum of the absolute values)
l2Norm Euclidean norm
l2NormSum Sum of Euclidean norms of variable blocks
nuclearNorm Nuclear norm (sum of the singular values)
elasticNet Elastic-net regularization function

Other functions

Name Description
quadratic Quadratic function with given Hessian and linear term
quadraticOverAffine Quadratic function subject to linear equality constraints
quadLoss Squared weighted Euclidean distance from a given vector
quadLossOverAffine Squared weighted Euclidean distance subject to linear equalities
logLoss Logistic loss function (used in logistic regression models for classification)
huberLoss Huber loss function (for robust least-squares regression)
hingeLoss Hinge loss function (used for example in SVM)
matFac Squared Frobenius norm of a matrix factorization residual

Linear operators

Linear operators composing the objective or the constraints can either be MATLAB’s matrices (dense or sparse) or constructed using the functions here listed. Additionally, operators from the comprehensive Spot toolbox can be readily used in defining problems, as they override standard MATLAB notation for handling matrices.

Name Description
linOp Constructs an operator given a linear map and its adjoint in the form of anonymous functions
hankelOp Operator computing the Hankel matrix of a given size associated with a vector