aGrUM 3.3.2
a C++ library for (probabilistic) graphical models
gum::MDDOperatorStrategy< GUM_ELEMENT > Class Template Reference

<agrum/FMDP/planning/mddOperatorStrategy.h> More...

#include <mddOperatorStrategy.h>

Inheritance diagram for gum::MDDOperatorStrategy< GUM_ELEMENT >:
[legend]
Collaboration diagram for gum::MDDOperatorStrategy< GUM_ELEMENT >:
[legend]

Public Member Functions

Constructor & destructor.
 MDDOperatorStrategy ()
 Default constructor.
 ~MDDOperatorStrategy () override
 Default destructor.
MultiDimFunctionGraph< GUM_ELEMENT, ExactTerminalNodePolicy > * getFunctionInstance () override
MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > * getArgMaxFunctionInstance () override
MultiDimFunctionGraph< ActionSet, SetTerminalNodePolicy > * getAggregatorInstance () override

Protected Member Functions

bool shouldEleminateVar_ (const DiscreteVariable *v, const FMDP< GUM_ELEMENT > *fmdp)
 Indicates if whether or not given var is to be eliminated. Called by the evalQaction.
const DiscreteVariable * lastVar_ (const MultiDimFunctionGraph< GUM_ELEMENT > *function)
 Returns the last var in the var order for given graph function Called by the evalQaction.
MultiDimFunctionGraph< GUM_ELEMENT > * frameTransition_ (const DiscreteVariable *x, const DiscreteVariable *xp)
 Returns a new function graph P(x'|x) = [x' == x].
void deleteFunctionGraph_ (const MultiDimFunctionGraph< GUM_ELEMENT > *f1, const MultiDimFunctionGraph< GUM_ELEMENT > *f2, Idx del)
void deleteFunctionGraph_ (const MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > *f1, const MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > *f2, Idx del)
Graph Function Operations Methods
MultiDimFunctionGraph< GUM_ELEMENT > * regress (const MultiDimFunctionGraph< GUM_ELEMENT > *Vold, Idx actionId, const FMDP< GUM_ELEMENT > *fmdp, const gum::VariableSet &elVarSeq) override
 Computes Qaction for given actionid.
MultiDimFunctionGraph< GUM_ELEMENT > * maximize (const MultiDimFunctionGraph< GUM_ELEMENT > *f1, const MultiDimFunctionGraph< GUM_ELEMENT > *f2, Idx del=3) override
MultiDimFunctionGraph< GUM_ELEMENT > * minimize (const MultiDimFunctionGraph< GUM_ELEMENT > *f1, const MultiDimFunctionGraph< GUM_ELEMENT > *f2, Idx del=3) override
MultiDimFunctionGraph< GUM_ELEMENT > * multiply (const MultiDimFunctionGraph< GUM_ELEMENT > *f1, const MultiDimFunctionGraph< GUM_ELEMENT > *f2, Idx del=3) override
MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > * argmaximize (const MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > *f1, const MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > *f2, Idx del=3) override
MultiDimFunctionGraph< GUM_ELEMENT > * add (const MultiDimFunctionGraph< GUM_ELEMENT > *f1, const MultiDimFunctionGraph< GUM_ELEMENT > *f2, Idx del=1) override
 Adds reward to given function( whether a qAction or vFunction).
MultiDimFunctionGraph< GUM_ELEMENT > * subtract (const MultiDimFunctionGraph< GUM_ELEMENT > *f1, const MultiDimFunctionGraph< GUM_ELEMENT > *f2, Idx del=0) override
 Subtract current VFunction from old VFunction to see if threshold is reached or not.

Detailed Description

template<typename GUM_ELEMENT>
class gum::MDDOperatorStrategy< GUM_ELEMENT >

<agrum/FMDP/planning/mddOperatorStrategy.h>

A class to find optimal policy for a given FMDP.

Perform a SPUDD planning on given in parameter factored markov decision process

Definition at line 71 of file mddOperatorStrategy.h.

Constructor & Destructor Documentation

◆ MDDOperatorStrategy()

template<typename GUM_ELEMENT>
gum::MDDOperatorStrategy< GUM_ELEMENT >::MDDOperatorStrategy ( )

Default constructor.

Definition at line 89 of file mddOperatorStrategy_tpl.h.

89 {
91 }
<agrum/FMDP/planning/mddOperatorStrategy.h>
MDDOperatorStrategy()
Default constructor.

References MDDOperatorStrategy().

Referenced by MDDOperatorStrategy(), and ~MDDOperatorStrategy().

Here is the call graph for this function:
Here is the caller graph for this function:

◆ ~MDDOperatorStrategy()

template<typename GUM_ELEMENT>
gum::MDDOperatorStrategy< GUM_ELEMENT >::~MDDOperatorStrategy ( )
override

Default destructor.

Definition at line 97 of file mddOperatorStrategy_tpl.h.

References MDDOperatorStrategy().

Here is the call graph for this function:

Member Function Documentation

◆ add()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< GUM_ELEMENT > * gum::MDDOperatorStrategy< GUM_ELEMENT >::add ( const MultiDimFunctionGraph< GUM_ELEMENT > * f1,
const MultiDimFunctionGraph< GUM_ELEMENT > * f2,
Idx del = 1 )
overrideprotectedvirtual

Adds reward to given function( whether a qAction or vFunction).

Warning
given function is deleted, returns the new one
Parameters
reward: R(s) or R(s,a)
function: either V(s) or Q(s,a)
Warning
given function is deleted, returns the new one

Implements gum::IOperatorStrategy< GUM_ELEMENT >.

Definition at line 213 of file mddOperatorStrategy_tpl.h.

215 {
217 this->deleteFunctionGraph_(f1, f2, del);
218 return ret;
219 }
void deleteFunctionGraph_(const MultiDimFunctionGraph< GUM_ELEMENT > *f1, const MultiDimFunctionGraph< GUM_ELEMENT > *f2, Idx del)

References gum::add2MultiDimFunctionGraphs(), and gum::IOperatorStrategy< GUM_ELEMENT >::deleteFunctionGraph_().

Here is the call graph for this function:

◆ argmaximize()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > * gum::MDDOperatorStrategy< GUM_ELEMENT >::argmaximize ( const MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > * f1,
const MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > * f2,
Idx del = 3 )
overrideprotectedvirtual
Warning
given vFunction and qAction are deleted, returns the new one

Implements gum::IOperatorStrategy< GUM_ELEMENT >.

Definition at line 191 of file mddOperatorStrategy_tpl.h.

References gum::MultiDimFunctionGraphOperator< GUM_ELEMENT, FUNCTOR, TerminalNodePolicy >::compute(), and gum::IOperatorStrategy< GUM_ELEMENT >::deleteFunctionGraph_().

Here is the call graph for this function:

◆ deleteFunctionGraph_() [1/2]

template<typename GUM_ELEMENT>
void gum::IOperatorStrategy< GUM_ELEMENT >::deleteFunctionGraph_ ( const MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > * f1,
const MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > * f2,
Idx del )
protectedinherited

Definition at line 78 of file IOperatorStrategy_tpl.h.

81 {
82 if (del == 1 || del == 3) delete f1;
83 if (del >= 2) delete f2;
84 }
<agrum/FMDP/SDyna/IOperatorStrategy.h>

◆ deleteFunctionGraph_() [2/2]

◆ frameTransition_()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< GUM_ELEMENT > * gum::IOperatorStrategy< GUM_ELEMENT >::frameTransition_ ( const DiscreteVariable * x,
const DiscreteVariable * xp )
protectedinherited

Returns a new function graph P(x'|x) = [x' == x].

It is the transition of a frame variable x, i.e., a variable with no transition for a given action: it keeps its value. The caller owns the returned function graph.

Parameters
xthe main variable
xpits primed version

Definition at line 49 of file IOperatorStrategy_tpl.h.

50 {
52 transition->add(*x);
53 transition->add(*xp);
54 auto* manager = transition->manager();
55 const auto root = manager->addInternalNode(x);
56 manager->setRootNode(root);
57 const auto zero = manager->addTerminalNode(GUM_ELEMENT(0));
58 const auto one = manager->addTerminalNode(GUM_ELEMENT(1));
59 for (Idx i = 0; i < x->domainSize(); ++i) {
60 const auto node = manager->addInternalNode(xp);
61 manager->setSon(root, i, node);
62 for (Idx ip = 0; ip < xp->domainSize(); ++ip)
63 manager->setSon(node, ip, (i == ip) ? one : zero);
64 }
65 return transition;
66 }
virtual MultiDimFunctionGraph< GUM_ELEMENT > * add(const MultiDimFunctionGraph< GUM_ELEMENT > *f1, const MultiDimFunctionGraph< GUM_ELEMENT > *f2, Idx del=1)=0
virtual MultiDimFunctionGraph< GUM_ELEMENT, ExactTerminalNodePolicy > * getFunctionInstance()=0

References gum::MultiDimFunctionGraph< GUM_ELEMENT, TerminalNodePolicy >::add(), gum::DiscreteVariable::domainSize(), getFunctionInstance(), and gum::MultiDimFunctionGraph< GUM_ELEMENT, TerminalNodePolicy >::manager().

Referenced by gum::MDDOperatorStrategy< GUM_ELEMENT >::regress(), gum::TreeOperatorStrategy< GUM_ELEMENT >::regress(), and subtract().

Here is the call graph for this function:
Here is the caller graph for this function:

◆ getAggregatorInstance()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< ActionSet, SetTerminalNodePolicy > * gum::MDDOperatorStrategy< GUM_ELEMENT >::getAggregatorInstance ( )
overridevirtual

◆ getArgMaxFunctionInstance()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< ArgMaxSet< GUM_ELEMENT, Idx >, SetTerminalNodePolicy > * gum::MDDOperatorStrategy< GUM_ELEMENT >::getArgMaxFunctionInstance ( )
overridevirtual

◆ getFunctionInstance()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< GUM_ELEMENT, ExactTerminalNodePolicy > * gum::MDDOperatorStrategy< GUM_ELEMENT >::getFunctionInstance ( )
overridevirtual

Implements gum::IOperatorStrategy< GUM_ELEMENT >.

Definition at line 253 of file mddOperatorStrategy_tpl.h.

253 {
255 }
static MultiDimFunctionGraph< GUM_ELEMENT, TerminalNodePolicy > * getReducedAndOrderedInstance()
Returns a reduced and ordered instance.

References gum::MultiDimFunctionGraph< GUM_ELEMENT, TerminalNodePolicy >::getReducedAndOrderedInstance().

Here is the call graph for this function:

◆ lastVar_()

template<typename GUM_ELEMENT>
const DiscreteVariable * gum::MDDOperatorStrategy< GUM_ELEMENT >::lastVar_ ( const MultiDimFunctionGraph< GUM_ELEMENT > * function)
protected

Returns the last var in the var order for given graph function Called by the evalQaction.

Definition at line 244 of file mddOperatorStrategy_tpl.h.

245 {
246 return function->variablesSequence().size() == 0
247 ? nullptr
248 : function->variablesSequence().atPos(function->variablesSequence().size() - 1);
249 }

References gum::MultiDimImplementation< GUM_ELEMENT >::variablesSequence().

Referenced by regress().

Here is the call graph for this function:
Here is the caller graph for this function:

◆ maximize()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< GUM_ELEMENT > * gum::MDDOperatorStrategy< GUM_ELEMENT >::maximize ( const MultiDimFunctionGraph< GUM_ELEMENT > * f1,
const MultiDimFunctionGraph< GUM_ELEMENT > * f2,
Idx del = 3 )
overrideprotectedvirtual
Warning
given vFunction and qAction are deleted, returns the new one

Implements gum::IOperatorStrategy< GUM_ELEMENT >.

Definition at line 155 of file mddOperatorStrategy_tpl.h.

References gum::IOperatorStrategy< GUM_ELEMENT >::deleteFunctionGraph_().

Here is the call graph for this function:

◆ minimize()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< GUM_ELEMENT > * gum::MDDOperatorStrategy< GUM_ELEMENT >::minimize ( const MultiDimFunctionGraph< GUM_ELEMENT > * f1,
const MultiDimFunctionGraph< GUM_ELEMENT > * f2,
Idx del = 3 )
overrideprotectedvirtual
Warning
Minimze the two given functions

Implements gum::IOperatorStrategy< GUM_ELEMENT >.

Definition at line 167 of file mddOperatorStrategy_tpl.h.

References gum::IOperatorStrategy< GUM_ELEMENT >::deleteFunctionGraph_().

Here is the call graph for this function:

◆ multiply()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< GUM_ELEMENT > * gum::MDDOperatorStrategy< GUM_ELEMENT >::multiply ( const MultiDimFunctionGraph< GUM_ELEMENT > * f1,
const MultiDimFunctionGraph< GUM_ELEMENT > * f2,
Idx del = 3 )
overrideprotectedvirtual
Warning
given f1 and f2 are deleted, returns the new one

Implements gum::IOperatorStrategy< GUM_ELEMENT >.

Definition at line 179 of file mddOperatorStrategy_tpl.h.

References gum::IOperatorStrategy< GUM_ELEMENT >::deleteFunctionGraph_(), and gum::multiply2MultiDimFunctionGraphs().

Here is the call graph for this function:

◆ regress()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< GUM_ELEMENT > * gum::MDDOperatorStrategy< GUM_ELEMENT >::regress ( const MultiDimFunctionGraph< GUM_ELEMENT > * Vold,
Idx actionId,
const FMDP< GUM_ELEMENT > * fmdp,
const gum::VariableSet & elVarSeq )
overrideprotectedvirtual

Computes Qaction for given actionid.

Performs a multiplication/projection on given qAction.

Parameters
qAction: the computed Q(s,a)
pxip: the transition probabilities for the xip variable
xip: the variable we eliminate on the projection
Warning
given qAction is deleted, return the new one

Implements gum::IOperatorStrategy< GUM_ELEMENT >.

Definition at line 118 of file mddOperatorStrategy_tpl.h.

121 {
124 qAction->copy(*Vold);
125
126 const DiscreteVariable* xip = this->lastVar_(qAction);
127
128 while (this->shouldEleminateVar_(xip, fmdp)) {
129 const DiscreteVariable* xi = fmdp->mapMainPrime().first(xip);
131 // a frame variable (without transition) keeps its value
133 if (pxi == nullptr) {
134 frame.reset(this->frameTransition_(xi, xip));
135 pxi = frame.get();
136 }
138 pxi,
139 &elVarSeq,
140 xip,
141 (GUM_ELEMENT)0);
143 delete qAction;
144 qAction = temp;
145 xip = this->lastVar_(qAction);
146 }
147
148 return qAction;
149 }
MultiDimFunctionGraph< GUM_ELEMENT > * frameTransition_(const DiscreteVariable *x, const DiscreteVariable *xp)
Returns a new function graph P(x'|x) = [x' == x].
const DiscreteVariable * lastVar_(const MultiDimFunctionGraph< GUM_ELEMENT > *function)
Returns the last var in the var order for given graph function Called by the evalQaction.
bool shouldEleminateVar_(const DiscreteVariable *v, const FMDP< GUM_ELEMENT > *fmdp)
Indicates if whether or not given var is to be eliminated. Called by the evalQaction.

References gum::Regress< GUM_ELEMENT, COMBINEOPERATOR, PROJECTOPERATOR, TerminalNodePolicy >::compute(), gum::MultiDimFunctionGraph< GUM_ELEMENT, TerminalNodePolicy >::copy(), gum::BijectionImplementation< T1, T2, std::is_scalar< T1 >::value &&std::is_scalar< T2 >::value >::first(), gum::IOperatorStrategy< GUM_ELEMENT >::frameTransition_(), gum::MultiDimFunctionGraph< GUM_ELEMENT, TerminalNodePolicy >::getReducedAndOrderedInstance(), lastVar_(), gum::FMDP< GUM_ELEMENT >::mapMainPrime(), RECAST, shouldEleminateVar_(), and gum::FMDP< GUM_ELEMENT >::transition().

Here is the call graph for this function:

◆ shouldEleminateVar_()

template<typename GUM_ELEMENT>
bool gum::MDDOperatorStrategy< GUM_ELEMENT >::shouldEleminateVar_ ( const DiscreteVariable * v,
const FMDP< GUM_ELEMENT > * fmdp )
protected

Indicates if whether or not given var is to be eliminated. Called by the evalQaction.

Definition at line 238 of file mddOperatorStrategy_tpl.h.

239 {
240 return v == nullptr ? false : fmdp->mapMainPrime().existsSecond(v);
241 }

References gum::BijectionImplementation< T1, T2, std::is_scalar< T1 >::value &&std::is_scalar< T2 >::value >::existsSecond(), and gum::FMDP< GUM_ELEMENT >::mapMainPrime().

Referenced by regress().

Here is the call graph for this function:
Here is the caller graph for this function:

◆ subtract()

template<typename GUM_ELEMENT>
MultiDimFunctionGraph< GUM_ELEMENT > * gum::MDDOperatorStrategy< GUM_ELEMENT >::subtract ( const MultiDimFunctionGraph< GUM_ELEMENT > * f1,
const MultiDimFunctionGraph< GUM_ELEMENT > * f2,
Idx del = 0 )
overrideprotectedvirtual

Subtract current VFunction from old VFunction to see if threshold is reached or not.

Warning
this time, nothing is deleted
Parameters
oldand new VFuntion
Warning
this time, nothing is deleted

Implements gum::IOperatorStrategy< GUM_ELEMENT >.

Definition at line 229 of file mddOperatorStrategy_tpl.h.

References gum::IOperatorStrategy< GUM_ELEMENT >::deleteFunctionGraph_(), and gum::subtract2MultiDimFunctionGraphs().

Here is the call graph for this function:

The documentation for this class was generated from the following files: