1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
|
// Copyright (C) 2025 EDF
// All Rights Reserved
// This code is published under the GNU Lesser General Public License (GNU LGPL)
#ifndef OPTIMIZERDPCUTGRIDADAPTBASE_H
#define OPTIMIZERDPCUTGRIDADAPTBASE_H
#include <memory>
#include <Eigen/Dense>
#include "StOpt/core/utils/StateWithStocks.h"
#include "StOpt/core/grids/GridAdaptBase.h"
#include "StOpt/regression/BaseRegression.h"
#include "StOpt/regression/ContinuationCutsGridAdaptNonConcave.h"
#include "StOpt/dp/SimulatorDPBase.h"
#include "StOpt/dp/OptimizerBase.h"
/** \file OptimizerDPCutGridAdaptBase.h
* \brief Define an abstract class for Dynamic Programming problems solved by regression methods using cust to approximate
* Bellman values
* Special version for stocks, concavity non supposed (it is imposed by mesh). Adaptation of the grid provided.
* \author Xavier Warin
*/
namespace StOpt
{
/// \class OptimizerDPCutGridAdaptBase OptimizerDPCutGridAdaptBase.h
/// Base class for optimizer for Dynamic Programming with regression methods and cuts, so using LP to solve transitional problems
class OptimizerDPCutGridAdaptBase : public OptimizerBase
{
public :
OptimizerDPCutGridAdaptBase() {}
virtual ~OptimizerDPCutGridAdaptBase() {}
/// \brief defines the dimension to split for MPI parallelism
/// For each dimension return true is the direction can be split
virtual Eigen::Array< bool, Eigen::Dynamic, 1> getDimensionToSplit() const = 0 ;
/// \brief defines a step in optimization
/// \param p_stock coordinates of the stock point to treat
/// \param p_condEsp continuation values
/// \return For a given simulation , cuts components (C) at a point stock \$ \bar S \f$ are given such that the cut is given by
/// \f$ C[0] + \sum_{i=1}^d C[i] (S_i - \bat S_i) \f$
virtual Eigen::ArrayXd stepOptimize(const Eigen::ArrayXd &p_stock, const StOpt::ContinuationCutsGridAdaptNonConcave &p_condEsp) const = 0;
/// \brief defines a step in simulation
/// Control are recalculated during simulation using a local optimzation using the LP
/// \param p_continuation defines the continuation operator
/// \param p_state defines the state value (modified)
/// \param p_phiInOut defines the value functions (modified) : size number of functions to follow
virtual void stepSimulate(const StOpt::ContinuationCutsGridAdaptNonConcave &p_continuation,
StOpt::StateWithStocks &p_state,
Eigen::Ref<Eigen::ArrayXd> p_phiInOut) const = 0 ;
/// \brief define a refinement of the grid to get a given precision
/// \param p_gridToAdapt grid to be adapted
/// \return True if a refinement has been achieved in the grid
virtual bool refineGrid( std::shared_ptr<GridAdaptBase> & p_gridToAdapt, const Eigen::ArrayXXd & pCuts ) =0;
// virtual void setRefinementToFalse()=0;
/// \brief get the simulator back
virtual std::shared_ptr< StOpt::SimulatorDPBase > getSimulator() const = 0;
/// \brief get back the dimension of the control
virtual int getNbControl() const { return 0;}
/// \brief get size of the function to follow in simulation
virtual int getSimuFuncSize() const = 0;
};
}
#endif /* OPTIMIZERDPCUTBASE_H */
|
