Package edu.pitt.csb.mgm

Class ConvexProximal

java.lang.Object

edu.pitt.csb.mgm.ConvexProximal

Direct Known Subclasses:

Mgm

public abstract class ConvexProximal
extends Object

This interface should be used for non-differentiable convex functions that are decomposable such that f(x) = g(x) + h(x) where g(x) is a differentiable convex function (i.e. smooth) and h(x) is a convex but not necessarily differentiable (i.e. non-smooth) and has a proximal operator prox_t(x) = argmin_z 1/(2t) norm2(x-z)^2 + h(z) has a solution for any t > 0. Typically g(x) will be a likelihood, and h(x) is a penalty term (as in l_1 in the lasso)

Created by ajsedgewick on 8/4/15.

Constructor Summary

Constructors

Constructor	Description
ConvexProximal()

Method Summary

Modifier and Type	Method	Description
double	nonSmooth(double t, cern.colt.matrix.DoubleMatrix1D X, cern.colt.matrix.DoubleMatrix1D Xout)	Calculate value of h(X) and proxOperator of h(X) at the same time for efficiency reasons.
double	smooth(cern.colt.matrix.DoubleMatrix1D X, cern.colt.matrix.DoubleMatrix1D Xout)	Calculate value of g(X) and gradient of g(X) at the same time for efficiency reasons.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

ConvexProximal

public ConvexProximal()

Method Details

smooth

public double smooth(cern.colt.matrix.DoubleMatrix1D X,
 cern.colt.matrix.DoubleMatrix1D Xout)

Calculate value of g(X) and gradient of g(X) at the same time for efficiency reasons.

Parameters:

X - input Vector

Xout - gradient of g(X)

Returns:

value of g(X)

nonSmooth

public double nonSmooth(double t,
 cern.colt.matrix.DoubleMatrix1D X,
 cern.colt.matrix.DoubleMatrix1D Xout)

Calculate value of h(X) and proxOperator of h(X) at the same time for efficiency reasons.

Parameters:

t - positive parameter for prox operator

X - input vector

Xout - vector solution to prox_t(X)

Returns:

value of h(X)