QPsolve: A wrapper to call solve.QP.

In scdensity: Shape-Constrained Kernel Density Estimation

Description

This function takes arguments slightly differently from solve.QP, to make it more convenient for internal use. It also implements measures to robustify calls to solve.QP:

• A rounding hack to prevent a bug in solve.QP that occasionally produces all-NaN solutions without returning a warning or error. Rounding has been found to eliminate almost all such bugs.

• A call to lpSolve's lp() to check feasibility before running solve.QP

• solve.QP is called within tryCatch to eliminate unwanted crashes.

The output of this function is a list with elements

• flag is 0 for successful completion, 1 for failure at the LP check stage, and 2 for failure at the QP stage (usually the "NaN solution" bug).

• QP is the list returned by solve.QP. If the QP was not run due to infeasibility, this element is NULL.

Usage

QPsolve(D, d, A, b, Aeq, beq)

Arguments

D	The matrix of the quadratic objective.
d	The vector in the linear term of the quadratic objective.
A	The matrix of inequality constraints.
b	The vector of RHS of the inequalities.
Aeq	The matrix of equality constraints.
beq	The vector of RHS of the equalities.

Details

solve.QP defines its quadratic program as minimizing 1/2 * x'Dx - x'd, subject to constraints A'x >= b. Equality constraints have to be in the first rows of A'.

This function minimizes x'Dx - x'd, subject to inequality constraints Ax >= b and Equality constraints Aeq*x = beq.

Value

A list with elements described above.

scdensity documentation built on May 1, 2019, 10:26 p.m.