ROL_ExpectationQuadRisk.hpp

Go to the documentation of this file.

// @HEADER
// *****************************************************************************
//               Rapid Optimization Library (ROL) Package
//
// Copyright 2014 NTESS and the ROL contributors.
// SPDX-License-Identifier: BSD-3-Clause
// *****************************************************************************
// @HEADER
 
#ifndef ROL_EXPECTATIONQUADRISK_HPP
#define ROL_EXPECTATIONQUADRISK_HPP
 
#include "ROL_ExpectationQuad.hpp"
#include "ROL_RandVarFunctional.hpp"
#include "ROL_Types.hpp"
 
namespace ROL {
 
template<class Real>
class ExpectationQuadRisk : public RandVarFunctional<Real> {
private:
  Ptr<ExpectationQuad<Real>> eq_;
 
  using RandVarFunctional<Real>::val_;
  using RandVarFunctional<Real>::gv_;
  using RandVarFunctional<Real>::g_;
  using RandVarFunctional<Real>::hv_;
  using RandVarFunctional<Real>::dualVector_;
 
  using RandVarFunctional<Real>::point_;
  using RandVarFunctional<Real>::weight_;
 
  using RandVarFunctional<Real>::computeValue;
  using RandVarFunctional<Real>::computeGradient;
  using RandVarFunctional<Real>::computeGradVec;
  using RandVarFunctional<Real>::computeHessVec;
 
public:
  ExpectationQuadRisk(const Ptr<ExpectationQuad<Real>> & eq)
    : RandVarFunctional<Real>(), eq_(eq) {}
 
  void checkRegret(void) {
    eq_->check();
  }
 
  void updateValue(Objective<Real>         &obj,
                   const Vector<Real>      &x,
                   const std::vector<Real> &xstat,
                   Real                    &tol) {
    Real val = computeValue(obj,x,tol);
    Real r   = eq_->regret(val-xstat[0],0);
    val_ += weight_ * r;
  }
 
  void updateGradient(Objective<Real>         &obj,
                      const Vector<Real>      &x,
                      const std::vector<Real> &xstat,
                      Real                    &tol) {
    Real val = computeValue(obj,x,tol);
    Real r   = eq_->regret(val-xstat[0],1);
    if (std::abs(r) >= ROL_EPSILON<Real>()) {
      val_ -= weight_ * r;
      computeGradient(*dualVector_,obj,x,tol);
      g_->axpy(weight_*r,*dualVector_);
    }
  }
 
  void updateHessVec(Objective<Real>         &obj,
                     const Vector<Real>      &v,
                     const std::vector<Real> &vstat,
                     const Vector<Real>      &x,
                     const std::vector<Real> &xstat,
                     Real                    &tol) {
    Real val = computeValue(obj,x,tol);
    Real r1  = eq_->regret(val-xstat[0],1);
    Real r2  = eq_->regret(val-xstat[0],2);
    if (std::abs(r2) >= ROL_EPSILON<Real>()) {
      Real gv = computeGradVec(*dualVector_,obj,v,x,tol);
      val_ += weight_ * r2 * (vstat[0] - gv);
      hv_->axpy(weight_*r2*(gv-vstat[0]),*dualVector_);
    }
    if (std::abs(r1) >= ROL_EPSILON<Real>()) {
      computeHessVec(*dualVector_,obj,v,x,tol);
      hv_->axpy(weight_*r1,*dualVector_);
    }
  }
 
  Real getValue(const Vector<Real>      &x,
                const std::vector<Real> &xstat,
                SampleGenerator<Real>   &sampler) {
    Real val(0);
    sampler.sumAll(&val_,&val,1);
    val += xstat[0];
    return val;
  }
 
  void getGradient(Vector<Real>            &g,
                   std::vector<Real>       &gstat,
                   const Vector<Real>      &x,
                   const std::vector<Real> &xstat,
                   SampleGenerator<Real>   &sampler) {
    Real stat(0), one(1);
    sampler.sumAll(&val_,&stat,1);
    stat += one;
    gstat[0] = stat;
    sampler.sumAll(*g_,g);
  }
 
  void getHessVec(Vector<Real>            &hv,
                  std::vector<Real>       &hvstat,
                  const Vector<Real>      &v,
                  const std::vector<Real> &vstat,
                  const Vector<Real>      &x,
                  const std::vector<Real> &xstat,
                  SampleGenerator<Real>   &sampler) {
    Real stat(0);
    sampler.sumAll(&val_,&stat,1);
    hvstat[0] = stat;
    sampler.sumAll(*hv_,hv);
  }
};
 
}
 
#endif