gum::NodeDatabase< AttributeSelection, isScalar > Class Template Reference

<agrum/FMDP/learning/datastructure/nodeDatabase.h> More...

#include <nodeDatabase.h>

Collaboration diagram for gum::NodeDatabase< AttributeSelection, isScalar >:

Public Member Functions
Idx	effectif (Idx moda) const
Idx	valueDomain () const
Constructor & destructor.
	NodeDatabase (const Set< const DiscreteVariable * > *, const DiscreteVariable *=nullptr)
	Default constructor. More...
	~NodeDatabase ()
	Default destructor. More...
void *	operator new (size_t s)
	Allocators and Deallocators redefinition. More...
void	operator delete (void *p)
	Default constructor. More...
Variable Test Methods
INLINE bool	isTestRelevant (const DiscreteVariable *var) const
	Indicates wether or not, node has sufficient observation so that any statistic is relevant. More...
INLINE double	testValue (const DiscreteVariable *var) const
	Returns the performance of given variables according to selection criterion. More...
INLINE double	testOtherCriterion (const DiscreteVariable *var) const
	Returns the performance of given variables according to selection secondary criterion (to break ties) More...
Aggregation Methods
NodeDatabase< AttributeSelection, isScalar > &	operator+= (const NodeDatabase< AttributeSelection, isScalar > &src)
	Merges given NodeDatabase informations into current nDB. More...
const TestPolicy< ValueType > *	testPolicy (const DiscreteVariable *var) const
	Returns a reference to nDB test policy for given variable (so that test policy information can be merged too) More...
const HashTableConstIteratorSafe< ValueType, Idx >	cbeginValues () const
	Iterators on value count to recopy correctly its content. More...
const HashTableConstIteratorSafe< ValueType, Idx >	cendValues () const
	Merges given NodeDatabase informations into current nDB. More...

Observation handling methods
void	addObservation (const Observation *)
	Nb observation taken into account by this instance. More...
INLINE Idx	nbObservation () const
	Nb observation taken into account by this instance. More...
void	_addObservation_ (const Observation *, Int2Type< true >)
	Updates database with new observation. More...
void	_addObservation_ (const Observation *, Int2Type< false >)
	depending on if we're learning reward function or transition probability More...

Detailed Description

template<TESTNAME AttributeSelection, bool isScalar>
class gum::NodeDatabase< AttributeSelection, isScalar >

<agrum/FMDP/learning/datastructure/nodeDatabase.h>

Definition at line 57 of file nodeDatabase.h.

Member Typedef Documentation

TestPolicy

template<TESTNAME AttributeSelection, bool isScalar>

template<typename GUM_SCALAR >

using gum::NodeDatabase< AttributeSelection, isScalar >::TestPolicy = typename TestSelect< AttributeSelection, GTestPolicy< GUM_SCALAR >, Chi2TestPolicy< GUM_SCALAR >, LeastSquareTestPolicy< GUM_SCALAR > >::type

private

Definition at line 64 of file nodeDatabase.h.

ValueType

template<TESTNAME AttributeSelection, bool isScalar>

typedef ValueSelect< isScalar, double, Idx >::type gum::NodeDatabase< AttributeSelection, isScalar >::ValueType

private

Definition at line 58 of file nodeDatabase.h.

Constructor & Destructor Documentation

NodeDatabase()

template<TESTNAME AttributeSelection, bool isScalar>

gum::NodeDatabase< AttributeSelection, isScalar >::NodeDatabase	(	const Set< const DiscreteVariable * > *	attrSet,
		const DiscreteVariable *	value = nullptr
	)

Default constructor.

Definition at line 42 of file nodeDatabase_tpl.h.

                                                  :
      _value_(value) {
    GUM_CONSTRUCTOR(NodeDatabase);

    for (SetIteratorSafe< const DiscreteVariable* > varIter = attrSet->cbeginSafe();
         varIter != attrSet->cendSafe();
         ++varIter)
      _attrTable_.insert(*varIter, new TestPolicy< ValueType >());

    _nbObservation_ = 0;
  }

~NodeDatabase()

template<TESTNAME AttributeSelection, bool isScalar>

gum::NodeDatabase< AttributeSelection, isScalar >::~NodeDatabase

(

)

Default destructor.

Definition at line 61 of file nodeDatabase_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                              {
    for (auto varIter = _attrTable_.beginSafe(); varIter != _attrTable_.endSafe(); ++varIter)
      delete varIter.val();

    GUM_DESTRUCTOR(NodeDatabase);
  }

Here is the call graph for this function:

Member Function Documentation

_addObservation_() [1/2]

template<TESTNAME AttributeSelection, bool isScalar>

void gum::NodeDatabase< AttributeSelection, isScalar >::_addObservation_ ( const Observation *  newObs, Int2Type< true >    )

private

Updates database with new observation.

Calls either or

Definition at line 88 of file nodeDatabase_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                        {
    for (auto varIter = _attrTable_.cbeginSafe(); varIter != _attrTable_.cendSafe(); ++varIter)
      varIter.val()->addObservation(newObs->rModality(varIter.key()), newObs->reward());

    if (_valueCount_.exists(newObs->reward()))
      _valueCount_[newObs->reward()]++;
    else
      _valueCount_.insert(newObs->reward(), 1);
  }

Here is the call graph for this function:

_addObservation_() [2/2]

template<TESTNAME AttributeSelection, bool isScalar>

void gum::NodeDatabase< AttributeSelection, isScalar >::_addObservation_ ( const Observation *  newObs, Int2Type< false >    )

private

depending on if we're learning reward function or transition probability

Definition at line 100 of file nodeDatabase_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                         {
    for (auto varIter = _attrTable_.cbeginSafe(); varIter != _attrTable_.cendSafe(); ++varIter)
      varIter.val()->addObservation(newObs->modality(varIter.key()), newObs->modality(_value_));

    if (_valueCount_.exists(newObs->modality(_value_)))
      _valueCount_[newObs->modality(_value_)]++;
    else
      _valueCount_.insert(newObs->modality(_value_), 1);
  }

Here is the call graph for this function:

_valueDomain_() [1/2]

template<TESTNAME AttributeSelection, bool isScalar>

Idx gum::NodeDatabase< AttributeSelection, isScalar >::_valueDomain_ ( Int2Type< true >  ) const

inlineprivate

Definition at line 190 of file nodeDatabase.h.

{ return _valueCount_.size(); }

_valueDomain_() [2/2]

template<TESTNAME AttributeSelection, bool isScalar>

Idx gum::NodeDatabase< AttributeSelection, isScalar >::_valueDomain_ ( Int2Type< false >  ) const

inlineprivate

Definition at line 191 of file nodeDatabase.h.

{ return _value_->domainSize(); }

addObservation()

template<TESTNAME AttributeSelection, bool isScalar>

void gum::NodeDatabase< AttributeSelection, isScalar >::addObservation

(

const Observation *

newObs

)

Nb observation taken into account by this instance.

Definition at line 82 of file nodeDatabase_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                                             {
    _nbObservation_++;
    this->_addObservation_(newObs, Int2Type< isScalar >());
  }

Here is the call graph for this function:

cbeginValues()

template<TESTNAME AttributeSelection, bool isScalar>

const HashTableConstIteratorSafe< ValueType, Idx > gum::NodeDatabase< AttributeSelection, isScalar >::cbeginValues ( ) const

inline

Iterators on value count to recopy correctly its content.

Definition at line 170 of file nodeDatabase.h.

                                                                            {
      return _valueCount_.cbeginSafe();
    }

cendValues()

template<TESTNAME AttributeSelection, bool isScalar>

const HashTableConstIteratorSafe< ValueType, Idx > gum::NodeDatabase< AttributeSelection, isScalar >::cendValues ( ) const

inline

Merges given NodeDatabase informations into current nDB.

Definition at line 173 of file nodeDatabase.h.

                                                                          {
      return _valueCount_.cendSafe();
    }

effectif()

template<TESTNAME AttributeSelection, bool isScalar>

Idx gum::NodeDatabase< AttributeSelection, isScalar >::effectif ( Idx  moda ) const

inline

Definition at line 183 of file nodeDatabase.h.

                                 {
      return _valueCount_.exists(ValueType(moda)) ? _valueCount_[ValueType(moda)] : 0;
    }

isTestRelevant()

template<TESTNAME AttributeSelection, bool isScalar>

INLINE bool gum::NodeDatabase< AttributeSelection, isScalar >::isTestRelevant ( const DiscreteVariable *  var ) const

inline

Indicates wether or not, node has sufficient observation so that any statistic is relevant.

Definition at line 128 of file nodeDatabase.h.

References gum::NodeDatabase< AttributeSelection, isScalar >::operator+=().

                                                                  {
      return _attrTable_[var]->isTestRelevant();
    }

Here is the call graph for this function:

nbObservation()

template<TESTNAME AttributeSelection, bool isScalar>

INLINE Idx gum::NodeDatabase< AttributeSelection, isScalar >::nbObservation ( ) const

inline

Nb observation taken into account by this instance.

Definition at line 115 of file nodeDatabase.h.

References gum::NodeDatabase< AttributeSelection, isScalar >::operator+=().

{ return _nbObservation_; }

Here is the call graph for this function:

operator delete()

template<TESTNAME AttributeSelection, bool isScalar>

void gum::NodeDatabase< AttributeSelection, isScalar >::operator delete ( void *  p )

inline

Default constructor.

Definition at line 86 of file nodeDatabase.h.

                                   {
      SmallObjectAllocator::instance().deallocate(p, sizeof(NodeDatabase));
    }

operator new()

template<TESTNAME AttributeSelection, bool isScalar>

void* gum::NodeDatabase< AttributeSelection, isScalar >::operator new ( size_t  s )

inline

Allocators and Deallocators redefinition.

Definition at line 85 of file nodeDatabase.h.

{ return SmallObjectAllocator::instance().allocate(s); }

operator+=()

template<TESTNAME AttributeSelection, bool isScalar>

NodeDatabase< AttributeSelection, isScalar > & gum::NodeDatabase< AttributeSelection, isScalar >::operator+=

(

const NodeDatabase< AttributeSelection, isScalar > &

src

)

Merges given NodeDatabase informations into current nDB.

Definition at line 122 of file nodeDatabase_tpl.h.

References gum::Set< Key, Alloc >::emplace().

Referenced by gum::NodeDatabase< AttributeSelection, isScalar >::isTestRelevant(), gum::NodeDatabase< AttributeSelection, isScalar >::nbObservation(), gum::NodeDatabase< AttributeSelection, isScalar >::testOtherCriterion(), and gum::NodeDatabase< AttributeSelection, isScalar >::testValue().

                                                                 {
    this->_nbObservation_ += src.nbObservation();

    for (auto varIter = _attrTable_.beginSafe(); varIter != _attrTable_.endSafe(); ++varIter)
      varIter.val()->add(*(src.testPolicy(varIter.key())));

    for (auto valIter = src.cbeginValues(); valIter != src.cendValues(); ++valIter)
      if (_valueCount_.exists(valIter.key()))
        _valueCount_[valIter.key()] += valIter.val();
      else
        _valueCount_.insert(valIter.key(), valIter.val());

    return *this;
  }

Here is the call graph for this function:

Here is the caller graph for this function:

testOtherCriterion()

template<TESTNAME AttributeSelection, bool isScalar>

INLINE double gum::NodeDatabase< AttributeSelection, isScalar >::testOtherCriterion ( const DiscreteVariable *  var ) const

inline

Returns the performance of given variables according to selection secondary criterion (to break ties)

Definition at line 142 of file nodeDatabase.h.

References gum::NodeDatabase< AttributeSelection, isScalar >::operator+=().

                                                                        {
      return _attrTable_[var]->secondaryscore();
    }

Here is the call graph for this function:

testPolicy()

template<TESTNAME AttributeSelection, bool isScalar>

const TestPolicy< ValueType >* gum::NodeDatabase< AttributeSelection, isScalar >::testPolicy ( const DiscreteVariable *  var ) const

inline

Returns a reference to nDB test policy for given variable (so that test policy information can be merged too)

Definition at line 163 of file nodeDatabase.h.

                                                                                 {
      return _attrTable_[var];
    }

testValue()

template<TESTNAME AttributeSelection, bool isScalar>

INLINE double gum::NodeDatabase< AttributeSelection, isScalar >::testValue ( const DiscreteVariable *  var ) const

inline

Returns the performance of given variables according to selection criterion.

Definition at line 136 of file nodeDatabase.h.

References gum::NodeDatabase< AttributeSelection, isScalar >::operator+=().

{ return _attrTable_[var]->score(); }

Here is the call graph for this function:

toString()

template<TESTNAME AttributeSelection, bool isScalar>

std::string gum::NodeDatabase< AttributeSelection, isScalar >::toString ( ) const

private

Definition at line 140 of file nodeDatabase_tpl.h.

References gum::Set< Key, Alloc >::emplace().

                                                                         {
    std::stringstream ss;

    ss << "NbObservation : " << this->nbObservation() << std::endl;
    for (auto varIter = _attrTable_.beginSafe(); varIter != _attrTable_.endSafe(); ++varIter)
      ss << "\t\tVariable : " << varIter.key()->name()
         << " - Associated Test : " << _attrTable_[varIter.key()]->toString() << std::endl;

    return ss.str();
  }

Here is the call graph for this function:

valueDomain()

template<TESTNAME AttributeSelection, bool isScalar>

Idx gum::NodeDatabase< AttributeSelection, isScalar >::valueDomain ( ) const

inline

Definition at line 187 of file nodeDatabase.h.

{ return _valueDomain_(Int2Type< isScalar >()); }

Member Data Documentation

_attrTable_

template<TESTNAME AttributeSelection, bool isScalar>

HashTable< const DiscreteVariable*, TestPolicy< ValueType >* > gum::NodeDatabase< AttributeSelection, isScalar >::_attrTable_

private

Table giving for every variables its instantiation.

Definition at line 197 of file nodeDatabase.h.

_nbObservation_

template<TESTNAME AttributeSelection, bool isScalar>

Idx gum::NodeDatabase< AttributeSelection, isScalar >::_nbObservation_

private

Definition at line 203 of file nodeDatabase.h.

_value_

template<TESTNAME AttributeSelection, bool isScalar>

const DiscreteVariable* gum::NodeDatabase< AttributeSelection, isScalar >::_value_

private

So does this reference on the value observed.

Definition at line 200 of file nodeDatabase.h.

_valueCount_

template<TESTNAME AttributeSelection, bool isScalar>

HashTable< ValueType, Idx > gum::NodeDatabase< AttributeSelection, isScalar >::_valueCount_

private

Definition at line 206 of file nodeDatabase.h.

---

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

• agrum/FMDP/learning/datastructure/nodeDatabase.h
• agrum/FMDP/learning/datastructure/nodeDatabase_tpl.h