gum::prm::StructuredBayesBall< GUM_SCALAR > Class Template Reference

<agrum/PRM/structuredBayesBall.h> More...

#include <structuredBayesBall.h>

Collaboration diagram for gum::prm::StructuredBayesBall< GUM_SCALAR >:

Public Member Functions
void	compute (const PRMInstance< GUM_SCALAR > *i, NodeId n)
	Compute the set or requisite nodes for each required instance given the current set of observations. Discard previous computations. More...
void	compute (const PRMInstance< GUM_SCALAR > &i, NodeId n)
	Compute the set or requisite nodes for each required instance given the current set of observations. Discard previous computations. More...
bool	isHardEvidence__ (const PRMInstance< GUM_SCALAR > *i, NodeId n)
	Returns true if there is a hard evidence on i->get(n). More...
Constructors & destructor.
	StructuredBayesBall (const PRMInference< GUM_SCALAR > &inference)
	Default Constructor. More...
	~StructuredBayesBall ()
	Destructor. More...
Getters and Setters.
const std::string &	key (const PRMInstance< GUM_SCALAR > *i) const
	Returns a unique key w.r.t. d-separation for i. More...
const std::string &	key (const PRMInstance< GUM_SCALAR > &i) const
	Returns a unique key w.r.t. d-separation for i. More...
const Set< NodeId > &	requisiteNodes (const PRMInstance< GUM_SCALAR > *i) const
	Returns the set of requisite nodes w.r.t. d-separation for i. More...
const Set< NodeId > &	requisiteNodes (const PRMInstance< GUM_SCALAR > &i) const
	Returns the set of requisite nodes w.r.t. d-separation for i. More...
Size	occurrence (const std::string &key) const
	Returns the number of occurrence of the given key, which is the number of PRMInstance<GUM_SCALAR> sharing the same set of requisite nodes. More...
float	liftRatio () const
	Returns the ratio between the total number of instances and the number of instances with the same configuration. More...
bool	exists (const PRMInstance< GUM_SCALAR > *i) const
	Returns true if i has requisite nodes. More...
bool	exists (const PRMInstance< GUM_SCALAR > &i) const
	Returns true if i has requisite nodes. More...

Detailed Description

template<typename GUM_SCALAR>
class gum::prm::StructuredBayesBall< GUM_SCALAR >

<agrum/PRM/structuredBayesBall.h>

This class represent the BayesBall algorithm applied on PRMs.

Definition at line 50 of file structuredBayesBall.h.

Member Typedef Documentation

InstanceMap

template<typename GUM_SCALAR >

typedef HashTable< const PRMInstance< GUM_SCALAR >*, MarkMap* > gum::prm::StructuredBayesBall< GUM_SCALAR >::InstanceMap

private

Code alias.

Definition at line 122 of file structuredBayesBall.h.

MarkMap

template<typename GUM_SCALAR >

typedef HashTable< NodeId, std::pair< bool, bool > > gum::prm::StructuredBayesBall< GUM_SCALAR >::MarkMap

private

Code alias.

Definition at line 120 of file structuredBayesBall.h.

Constructor & Destructor Documentation

StructuredBayesBall() [1/2]

template<typename GUM_SCALAR >

INLINE gum::prm::StructuredBayesBall< GUM_SCALAR >::StructuredBayesBall

(

const PRMInference< GUM_SCALAR > &

inference

)

Default Constructor.

Definition at line 270 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                    :
        inf__(&inference) {
      GUM_CONSTRUCTOR(StructuredBayesBall);
    }

Here is the call graph for this function:

~StructuredBayesBall()

template<typename GUM_SCALAR >

gum::prm::StructuredBayesBall< GUM_SCALAR >::~StructuredBayesBall

(

)

Destructor.

Definition at line 35 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                            {
      GUM_DESTRUCTOR(StructuredBayesBall);

      for (const auto& elt: reqMap__)
        delete elt.second.first;
    }

Here is the call graph for this function:

StructuredBayesBall() [2/2]

template<typename GUM_SCALAR >

gum::prm::StructuredBayesBall< GUM_SCALAR >::StructuredBayesBall ( const StructuredBayesBall< GUM_SCALAR > &  source )

private

Copy constructor.

Member Function Documentation

buildHashKey__()

template<typename GUM_SCALAR >

std::string gum::prm::StructuredBayesBall< GUM_SCALAR >::buildHashKey__ ( const PRMInstance< GUM_SCALAR > *  i, Set< NodeId > &  req_nodes  )

private

Builds the HashKey for the given instance and requisite nodes set.

Definition at line 257 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                   {
      std::stringstream sBuff;
      sBuff << i->type().name();

      for (const auto node: i->type().containerDag().nodes())
        if (req_nodes.exists(node)) sBuff << "-" << node;

      return sBuff.str();
    }

Here is the call graph for this function:

clean__()

template<typename GUM_SCALAR >

void gum::prm::StructuredBayesBall< GUM_SCALAR >::clean__ ( )

private

Cleans this before a new computation.

Definition at line 43 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                    {
      for (const auto& elt: reqMap__)
        delete elt.second.first;

      keyMap__.clear();
      reqMap__.clear();
    }

Here is the call graph for this function:

compute() [1/2]

template<typename GUM_SCALAR >

INLINE void gum::prm::StructuredBayesBall< GUM_SCALAR >::compute	(	const PRMInstance< GUM_SCALAR > *	i,
		NodeId	n
	)

Compute the set or requisite nodes for each required instance given the current set of observations. Discard previous computations.

Definition at line 339 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                           {
      compute__(i, n);
    }

Here is the call graph for this function:

compute() [2/2]

template<typename GUM_SCALAR >

INLINE void gum::prm::StructuredBayesBall< GUM_SCALAR >::compute	(	const PRMInstance< GUM_SCALAR > &	i,
		NodeId	n
	)

Compute the set or requisite nodes for each required instance given the current set of observations. Discard previous computations.

Definition at line 346 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                           {
      compute__(&i, n);
    }

Here is the call graph for this function:

compute__()

template<typename GUM_SCALAR >

void gum::prm::StructuredBayesBall< GUM_SCALAR >::compute__ ( const PRMInstance< GUM_SCALAR > *  i, NodeId  n  )

private

The real compute method.

Definition at line 79 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                           {
      clean__();
      StructuredBayesBall< GUM_SCALAR >::InstanceMap marks;
      fromChild__(i, n, marks);
      fillMaps__(marks);

      for (const auto& elt: marks)
        delete elt.second;
    }

Here is the call graph for this function:

exists() [1/2]

template<typename GUM_SCALAR >

INLINE bool gum::prm::StructuredBayesBall< GUM_SCALAR >::exists

(

const PRMInstance< GUM_SCALAR > *

i

)

const

Returns true if i has requisite nodes.

Definition at line 327 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                 {
      return keyMap__.exists(i);
    }

Here is the call graph for this function:

exists() [2/2]

template<typename GUM_SCALAR >

INLINE bool gum::prm::StructuredBayesBall< GUM_SCALAR >::exists

(

const PRMInstance< GUM_SCALAR > &

i

)

const

Returns true if i has requisite nodes.

Definition at line 333 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                 {
      return keyMap__.exists(&i);
    }

Here is the call graph for this function:

fillMaps__()

template<typename GUM_SCALAR >

void gum::prm::StructuredBayesBall< GUM_SCALAR >::fillMaps__ ( InstanceMap &  marks )

private

Fill keyMap__ and reqMap__.

Definition at line 212 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                                         {
      // First find for each instance it's requisite nodes
      HashTable< const PRMInstance< GUM_SCALAR >*, Set< NodeId >* > req_map;

      for (const auto& elt: marks) {
        Set< NodeId >* req_set = new Set< NodeId >();

        for (const auto& elt2: *elt.second)
          if (elt2.second.first) req_set->insert(elt2.first);

        req_map.insert(elt.first, req_set);
      }

      // Remove all instances with 0 requisite nodes
      Set< const PRMInstance< GUM_SCALAR >* > to_remove;

      for (const auto& elt: req_map)
        if (elt.second->size() == 0) to_remove.insert(elt.first);

      for (const auto remo: to_remove) {
        delete req_map[remo];
        req_map.erase(remo);
      }

      // Fill reqMap__ and keyMap__
      for (const auto& elt: req_map) {
        std::string key = buildHashKey__(elt.first, *elt.second);

        if (reqMap__.exists(key)) {
          keyMap__.insert(
             elt.first,
             std::pair< std::string, Set< NodeId >* >(key, reqMap__[key].first));
          reqMap__[key].second += 1;
          delete elt.second;
          req_map[elt.first] = 0;
        } else {
          reqMap__.insert(key, std::pair< Set< NodeId >*, Size >(elt.second, 1));
          keyMap__.insert(
             elt.first,
             std::pair< std::string, Set< NodeId >* >(key, elt.second));
        }
      }
    }

Here is the call graph for this function:

fromChild__()

template<typename GUM_SCALAR >

void gum::prm::StructuredBayesBall< GUM_SCALAR >::fromChild__ ( const PRMInstance< GUM_SCALAR > *  i, NodeId  n, InstanceMap &  marks  )

private

When the ball is received on i->get(n) from a child.

Definition at line 94 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                               {
      if (!marks.exists(i)) {
        marks.insert(i, new StructuredBayesBall< GUM_SCALAR >::MarkMap());
      }

      if (!marks[i]->exists(n)) {
        marks[i]->insert(n, std::pair< bool, bool >(false, false));
      }

      // Sending message to parents
      switch (i->type().get(n).elt_type()) {
        case PRMClassElement< GUM_SCALAR >::prm_slotchain: {
          if (!getMark__(marks, i, n).first) {
            getMark__(marks, i, n).first = true;

            for (const auto inst: i->getInstances(n))
              fromChild__(inst,
                          inst->get(getSC__(i, n).lastElt().safeName()).id(),
                          marks);
          }

          if (!getMark__(marks, i, n).second) {
            getMark__(marks, i, n).second = true;

            for (const auto chi: i->type().containerDag().children(n))
              fromParent__(i, chi, marks);
          }

          break;
        }

        case PRMClassElement< GUM_SCALAR >::prm_aggregate:
        case PRMClassElement< GUM_SCALAR >::prm_attribute: {
          if (!getMark__(marks, i, n).first) {
            getMark__(marks, i, n).first = true;

            if (!isHardEvidence__(i, n))
              for (const auto par: i->type().containerDag().parents(n))
                fromChild__(i, par, marks);
          }

          if (!getMark__(marks, i, n).second) {
            getMark__(marks, i, n).second = true;

            // In i.
            for (const auto chi: i->type().containerDag().children(n))
              fromParent__(i, chi, marks);

            // Out of i.
            try {
              const auto& refs = i->getRefAttr(n);

              for (auto iter = refs.begin(); iter != refs.end(); ++iter)
                fromParent__(iter->first,
                             iter->first->type().get(iter->second).id(),
                             marks);
            } catch (NotFound&) {
              // Not an inverse sc
            }
          }

          break;
        }

        default: {
          // We shouldn't reach any other PRMClassElement<GUM_DATA> than
          // PRMAttribute
          // or
          // PRMSlotChain<GUM_SCALAR>.
          GUM_ERROR(FatalError, "This case is impossible.");
        }
      }
    }

Here is the call graph for this function:

fromParent__()

template<typename GUM_SCALAR >

void gum::prm::StructuredBayesBall< GUM_SCALAR >::fromParent__ ( const PRMInstance< GUM_SCALAR > *  i, NodeId  n, InstanceMap &  marks  )

private

When the ball is receive on i->get(n) from a parent.

Definition at line 172 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                               {
      if (!marks.exists(i)) {
        marks.insert(i, new StructuredBayesBall< GUM_SCALAR >::MarkMap());
      }

      if (!marks[i]->exists(n)) {
        marks[i]->insert(n, std::pair< bool, bool >(false, false));
      }

      // Concerns only PRMAttribute (because of the hard evidence)
      if ((isHardEvidence__(i, n)) && (!getMark__(marks, i, n).first)) {
        getMark__(marks, i, n).first = true;

        for (const auto par: i->type().containerDag().parents(n))
          fromChild__(i, par, marks);
      } else if (!getMark__(marks, i, n).second) {
        getMark__(marks, i, n).second = true;

        // In i.
        for (const auto chi: i->type().containerDag().children(n))
          fromParent__(i, chi, marks);

        // Out of i.
        try {
          for (auto iter = i->getRefAttr(n).begin();
               iter != i->getRefAttr(n).end();
               ++iter)
            fromParent__(iter->first,
                         iter->first->type().get(iter->second).id(),
                         marks);
        } catch (NotFound&) {
          // Not an inverse sc
        }
      }
    }

Here is the call graph for this function:

getMark__()

template<typename GUM_SCALAR >

INLINE std::pair< bool, bool > & gum::prm::StructuredBayesBall< GUM_SCALAR >::getMark__ ( InstanceMap &  marks, const PRMInstance< GUM_SCALAR > *  i, NodeId  n  )

private

Code alias.

Definition at line 361 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                           {
      return (*(marks[i]))[n];
    }

Here is the call graph for this function:

getSC__()

template<typename GUM_SCALAR >

INLINE const PRMSlotChain< GUM_SCALAR > & gum::prm::StructuredBayesBall< GUM_SCALAR >::getSC__ ( const PRMInstance< GUM_SCALAR > *  i, NodeId  n  )

private

Code alias.

Definition at line 354 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                              {
      return static_cast< const PRMSlotChain< GUM_SCALAR >& >(i->type().get(n));
    }

Here is the call graph for this function:

isHardEvidence__()

template<typename GUM_SCALAR >

bool gum::prm::StructuredBayesBall< GUM_SCALAR >::isHardEvidence__	(	const PRMInstance< GUM_SCALAR > *	i,
		NodeId	n
	)

Returns true if there is a hard evidence on i->get(n).

Definition at line 52 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                           {
      try {
        typename PRMInference< GUM_SCALAR >::Chain chain
           = std::make_pair(i, &(i->get(n)));

        if (inf__->hasEvidence(chain)) {
          const Potential< GUM_SCALAR >* e = inf__->evidence(i)[n];
          Instantiation                  inst(e);
          Size                           count = 0;

          for (inst.setFirst(); !inst.end(); inst.inc()) {
            if ((e->get(inst) == (GUM_SCALAR)1.0))
              ++count;
            else if (e->get(inst) != (GUM_SCALAR)0.0)
              return false;
          }

          return (count == 1);
        }

        return false;
      } catch (NotFound&) { return false; }
    }

Here is the call graph for this function:

key() [1/2]

template<typename GUM_SCALAR >

INLINE const std::string & gum::prm::StructuredBayesBall< GUM_SCALAR >::key

(

const PRMInstance< GUM_SCALAR > *

i

)

const

Returns a unique key w.r.t. d-separation for i.

Definition at line 292 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                 {
      return keyMap__[i].first;
    }

Here is the call graph for this function:

key() [2/2]

template<typename GUM_SCALAR >

INLINE const std::string & gum::prm::StructuredBayesBall< GUM_SCALAR >::key

(

const PRMInstance< GUM_SCALAR > &

i

)

const

Returns a unique key w.r.t. d-separation for i.

Definition at line 298 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                 {
      return keyMap__[&i].first;
    }

Here is the call graph for this function:

liftRatio()

template<typename GUM_SCALAR >

INLINE float gum::prm::StructuredBayesBall< GUM_SCALAR >::liftRatio

(

)

const

Returns the ratio between the total number of instances and the number of instances with the same configuration.

Definition at line 322 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                                    {
      return ((float)reqMap__.size()) / ((float)keyMap__.size());
    }

Here is the call graph for this function:

occurrence()

template<typename GUM_SCALAR >

INLINE Size gum::prm::StructuredBayesBall< GUM_SCALAR >::occurrence

(

const std::string &

key

)

const

Returns the number of occurrence of the given key, which is the number of PRMInstance<GUM_SCALAR> sharing the same set of requisite nodes.

Definition at line 316 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                   {
      return reqMap__[key].second;
    }

Here is the call graph for this function:

operator=()

template<typename GUM_SCALAR >

INLINE StructuredBayesBall< GUM_SCALAR > & gum::prm::StructuredBayesBall< GUM_SCALAR >::operator= ( const StructuredBayesBall< GUM_SCALAR > &  source )

private

Copy operator.

Definition at line 286 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                           {
      GUM_ERROR(FatalError, "Not allowed.");
    }

Here is the call graph for this function:

requisiteNodes() [1/2]

template<typename GUM_SCALAR >

INLINE const Set< NodeId > & gum::prm::StructuredBayesBall< GUM_SCALAR >::requisiteNodes

(

const PRMInstance< GUM_SCALAR > *

i

)

const

Returns the set of requisite nodes w.r.t. d-separation for i.

Definition at line 304 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                 {
      return *(keyMap__[i].second);
    }

Here is the call graph for this function:

requisiteNodes() [2/2]

template<typename GUM_SCALAR >

INLINE const Set< NodeId > & gum::prm::StructuredBayesBall< GUM_SCALAR >::requisiteNodes

(

const PRMInstance< GUM_SCALAR > &

i

)

const

Returns the set of requisite nodes w.r.t. d-separation for i.

Definition at line 310 of file structuredBayesBall_tpl.h.

References gum::prm::ParamScopeData< GUM_SCALAR >::ParamScopeData().

                                                 {
      return *(keyMap__[&i].second);
    }

Here is the call graph for this function:

Member Data Documentation

inf__

template<typename GUM_SCALAR >

const PRMInference< GUM_SCALAR >* gum::prm::StructuredBayesBall< GUM_SCALAR >::inf__

private

The PRM at which model__ belongs.

Definition at line 155 of file structuredBayesBall.h.

keyMap__

template<typename GUM_SCALAR >

HashTable< const PRMInstance< GUM_SCALAR >*, std::pair< std::string, Set< NodeId >* > > gum::prm::StructuredBayesBall< GUM_SCALAR >::keyMap__

private

Associate an PRMInstance<GUM_SCALAR> with a unique key w.r.t. d-separation and the set of requisite nodes deduced from d-separation analysis.

Definition at line 163 of file structuredBayesBall.h.

reqMap__

template<typename GUM_SCALAR >

HashTable< std::string, std::pair< Set< NodeId >*, Size > > gum::prm::StructuredBayesBall< GUM_SCALAR >::reqMap__

private

Associate a Key with the set of requisite nodes associated with it. The Size value is the number of instance with the same key.

Definition at line 167 of file structuredBayesBall.h.

---

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

• agrum/PRM/inference/structuredBayesBall.h
• agrum/PRM/inference/structuredBayesBall_tpl.h