FactoredDecPOMDPDiscrete is implements a factored DecPOMDPDiscrete. More...

#include <FactoredDecPOMDPDiscrete.h>

Inheritance diagram for FactoredDecPOMDPDiscrete:

Public Member Functions
virtual void	CacheFlatModels (bool sparse)

void	CacheFlatRewardModel (bool sparse=false)

void	ClipRewardModel (Index sf, bool sparse)
	Adjusts the reward model by ignoring a state factor. More...

virtual FactoredDecPOMDPDiscrete *	Clone () const
	Returns a pointer to a copy of this class. More...

virtual double	ComputeObservationProb (Index o, Index oVal, const std::vector< Index > &As, const std::vector< Index > &Ys, const std::vector< Index > &Os) const

virtual double	ComputeTransitionProb (Index y, Index yVal, const std::vector< Index > &Xs, const std::vector< Index > &As, const std::vector< Index > &Ys) const

void	ConvertFiniteToInfiniteHorizon (size_t horizon)

void	CreateNewRewardModel ()
	Creates a new reward model mapping. More...

void	CreateNewRewardModelForAgent (Globals::Index)

virtual void	ExportSpuddFile (const std::string &filename) const
	Export fully-observable subset of this FactoredDecPOMDPDiscrete to spudd file. More...

	FactoredDecPOMDPDiscrete (std::string name="received unspec. by FactoredDecPOMDPDiscrete", std::string descr="received unspec. by FactoredDecPOMDPDiscrete", std::string pf="received unspec. by FactoredDecPOMDPDiscrete")
	Default constructor. More...

const Scope &	GetAgentScopeForLRF (Index LRF) const
	Get all the agents in the scope of an LRF. More...

const FactoredQFunctionScopeForStage	GetImmediateRewardScopes () const
	Returns all scopes of the immediate reward function in one object. More...

RewardModel *	GetLRF (Index LRF) const
	Returns a pointer to the LRF-th reward model component. More...

double	GetLRFReward (Index LRF, Index sI_e, Index jaI_e) const
	Returns reward for LRF, given ja_e - the joint index for a group action for the subset of agents as specified by the agent scope of LRF. More...

double	GetLRFReward (Index LRF, const std::vector< Index > &sI_e, const std::vector< Index > &jaI_e) const

double	GetLRFRewardFlat (Index LRF, Index flat_s, Index full_ja) const
	might be necessary? Returns reward for LRF, given a flat state index, and a full joint action. More...

double	GetLRFRewardFlat (Index lrf, const std::vector< Index > &sfacs, const std::vector< Index > &as) const

size_t	GetNrAIs (Index LRF) const
	Get the number of action instantiations. I. More...

size_t	GetNrLRFs () const
	Get the number of LRFs in the factored representation. More...

size_t	GetNrXIs (Index LRF) const
	Get the number of X instantiations for LRF. More...

double	GetReward (const std::vector< Index > &sIs, const std::vector< Index > &aIs) const

double	GetReward (Index sI, Index jaI) const
	Return the reward for state, joint action indices. More...

double	GetReward (State s, JointAction ja) const
	Function that returns the reward for a state and joint action. More...

double	GetRewardForAgent (Index agentI, Index sI, Index jaI) const
	Return the reward for state, joint action indices. More...

double	GetRewardForAgent (Index agI, State s, JointAction ja) const
	Function that returns the reward for a state and joint action. More...

RGet *	GetRGet () const

const Scope &	GetStateFactorScopeForLRF (Index LRF) const
	Get all the state factors in the scope of an LRF. More...

void	InitializeInstantiationInformation ()
	Initializes some meta information computed from the scopes. More...

void	InitializeStorage ()
	makes sure that _m_sfScopes, _m_agScopes, _m_LRFs are set to proper size More...

void	MarginalizeISD (Index sf, std::vector< size_t > &factor_sizes, const FactoredStateDistribution *fsd)
	Marginalizes the ISD over a given state factor. More...

void	MarginalizeStateFactor (Index sf, bool sparse)
	Attempts to remove an extraneous state factor from the problem. More...

void	Print () const

Index	RestrictedActionVectorToJointIndex (Index LRF, const std::vector< Index > &actionVec_e) const
	convert an action vector of restricted scope to a joint index a_e. More...

Index	RestrictedStateVectorToJointIndex (Index LRF, const std::vector< Index > &stateVec_e) const
	Get the vector of Factor indices corresponding to stateI. More...

void	SetNrLRFs (size_t nr)

virtual void	SetOScopes ()

void	SetReward (Index sI, Index jaI, double r)
	Set the reward for state, joint action indices. More...

void	SetReward (Index sI, Index jaI, Index sucSI, double r)
	Set the reward for state, joint action , suc. state indices. More...

void	SetReward (Index sI, Index jaI, Index sucSI, Index joI, double r)
	Set the reward for state, joint action, suc.state, joint obs indices. More...

void	SetReward (State s, JointAction ja, double r)
	Function that sets the reward for a state and joint action. More...

void	SetRewardForAgent (Index agentI, Index sI, Index jaI, double r)

void	SetRewardForAgent (Index agentI, Index sI, Index jaI, Index sucSI, double r)
	Set the reward for state, joint action , suc. state indices. More...

void	SetRewardForAgent (Index agentI, Index sI, Index jaI, Index sucSI, Index joI, double r)
	Set the reward for state, joint action, suc.state, joint obs indices. More...

void	SetRewardForAgent (Index agI, State s, JointAction ja, double r)
	Function that sets the reward for an agent, state and joint action. More...

void	SetRewardForLRF (Index LRF, const std::vector< Index > &Xs, const std::vector< Index > &As, double reward)
	Set reward. More...

void	SetRewardForLRF (Index LRF, const std::vector< Index > &Xs, const std::vector< Index > &As, const Scope &Y, const std::vector< Index > &Ys, const Scope &O, const std::vector< Index > &Os, double reward)
	Set reward. More...

void	SetRewardForLRF (Index LRF, const Scope &X, const std::vector< Index > &Xs, const Scope &A, const std::vector< Index > &As, const Scope &Y, const std::vector< Index > &Ys, const Scope &O, const std::vector< Index > &Os, double reward)
	Set Reward. More...

void	SetRM (Index LRF, RewardModel *rm)
	Add the LRF-th reward model. More...

void	SetScopeForLRF (Index LRF, const Scope &X, const Scope &A)
	Sets the scope for LRF. More...

void	SetScopeForLRF (Index LRF, const Scope &X, const Scope &A, const Scope &Y, const Scope &O)
	Sets the scope for LRF. More...

virtual void	SetYScopes ()

virtual std::string	SoftPrint () const
	Prints some information on the MultiAgentDecisionProcessDiscreteInterface. More...

virtual	~FactoredDecPOMDPDiscrete ()
	Destructor. More...

Public Member Functions inherited from FactoredDecPOMDPDiscreteInterface
virtual	~FactoredDecPOMDPDiscreteInterface ()
	(default) Constructor More...

Public Member Functions inherited from DecPOMDPDiscreteInterface
virtual	~DecPOMDPDiscreteInterface ()
	import the GetReward function from the base class in current scope. More...

Public Member Functions inherited from DecPOMDPInterface
virtual	~DecPOMDPInterface ()
	Virtual destructor. More...

Public Member Functions inherited from POSGInterface
virtual	~POSGInterface ()
	Virtual destructor. More...

Public Member Functions inherited from MultiAgentDecisionProcessInterface
virtual	~MultiAgentDecisionProcessInterface ()
	Destructor. More...

Public Member Functions inherited from MultiAgentDecisionProcessDiscreteInterface
virtual	~MultiAgentDecisionProcessDiscreteInterface ()
	Destructor. More...

Public Member Functions inherited from MultiAgentDecisionProcessDiscreteFactoredStatesInterface
std::string	SoftPrintState (Index sI) const
	SoftPrints information on the MultiAgentDecisionProcessDiscrete. More...

virtual	~MultiAgentDecisionProcessDiscreteFactoredStatesInterface ()
	(default) Constructor More...

Public Member Functions inherited from MultiAgentDecisionProcessDiscreteFactoredStates
void	AddAction (Index AI, const std::string &name, const std::string &description="")

void	AddObservation (Index AI, const std::string &name, const std::string &description="")

Index	AddStateFactor (const std::string &n="undef. name", const std::string &d="undef. descr.")

Index	AddStateFactorValue (Index sf, const std::string &v="undef")

Scope	AgentScopeBackup (const Scope &stateScope, const Scope &agentScope) const

void	CacheFlatObservationModel (bool sparse=false)

void	CacheFlatTransitionModel (bool sparse=false)

size_t	ConstructJointActions ()

size_t	ConstructJointObservations ()

void	CreateNewObservationModel ()
	Creates a new observation model mapping: initializes new CPDs. More...

void	CreateNewTransitionModel ()
	Creates a new transition model: initializes new CPDs for the 2BDN. More...

Index	FactorValueIndicesToStateIndex (const std::vector< Index > &fv) const
	convert std::vector of (indices of) factor values to (flat) state index. More...

Index	FactorValueIndicesToStateIndex (const std::vector< Index > &s_e_vec, const Scope &sfSC) const
	convert an local state vector s_e_vec of scope sfScope to a joint index. More...

const TwoStageDynamicBayesianNetwork *	Get2DBN () const
	Copy assignment operator. More...

const Action *	GetAction (Index agentI, Index a) const
	Return a ref to the a-th action of agent agentI. More...

const Scope &	GetAllStateFactorScope () const
	Convenience function to quickly get the full state scope. More...

const Scope &	GetASoI_O (Index o) const

const Scope &	GetASoI_Y (Index y) const

bool	GetEventObservability () const
	Are we using an event observation model? More...

const FactoredStateDistribution *	GetFactoredISD () const

double	GetInitialStateProbability (Index sI) const

StateDistribution *	GetISD () const
	Returns the complete initial state distribution. More...

const JointAction *	GetJointAction (Index i) const
	Return a ref to the i-th joint action. More...

const JointObservation *	GetJointObservation (Index i) const
	Return a ref to the i-th joint observation. More...

const std::vector< size_t > &	GetNrActions () const
	Return the number of actions vector. More...

size_t	GetNrActions (Index AgentI) const
	Return the number of actions of agent agentI. More...

size_t	GetNrJointActions () const
	Return the number of joint actions. More...

size_t	GetNrJointActions (const Scope &agScope) const
	Get the number of joint actions the agents in agScope can form. More...

size_t	GetNrJointObservations () const
	Return the number of joint observations. More...

const std::vector< size_t > &	GetNrObservations () const
	Return the number of observations vector. More...

size_t	GetNrObservations (Index AgentI) const
	Return the number of observations of agent agentI. More...

size_t	GetNrStateFactorInstantiations (const Scope &sfScope) const
	Get the number of joint instantiations for the factors in sfScope. More...

size_t	GetNrStateFactors () const
	Get the number of state components. -no is inherited from. More...

size_t	GetNrStates () const
	Return the number of states. More...

const size_t	GetNrValuesForFactor (Index sf) const
	Get the number of possible values for a particular factor. More...

const std::vector< size_t > &	GetNrValuesPerFactor () const
	Get the number of possible assignments or values to each factor. More...

const Observation *	GetObservation (Index agentI, Index a) const
	Return a ref to the a-th observation of agent agentI. More...

const ObservationModelDiscrete *	GetObservationModelDiscretePtr () const
	Returns a pointer to the underlying observation model. More...

double	GetObservationProbability (Index jaI, Index sucSI, Index joI) const
	Return the probability of joint observation joI: P(joI\|jaI,sucSI). More...

double	GetObservationProbability (Index sI, Index jaI, Index sucSI, Index joI) const
	O(s,ja,s',jo) version. You can access a standard O(ja,s',jo) model both ways (the PS index is simply ignored in the latter case). More...

OGet *	GetOGet () const

const Scope &	GetOSoI_O (Index o) const

const State *	GetState (Index i) const
	Returns a pointer to state i. More...

const StateFactorDiscrete *	GetStateFactorDiscrete (Index sfacI) const

TGet *	GetTGet () const

const TransitionModelDiscrete *	GetTransitionModelDiscretePtr () const
	Returns a pointer to the underlying transition model. More...

double	GetTransitionProbability (Index sI, Index jaI, Index sucSI) const
	Return the probability of successor state sucSI: P(sucSI\|sI,jaI). More...

const Scope &	GetXSoI_O (Index o) const

const Scope &	GetXSoI_Y (Index y) const

double	GetYOProbability (const Scope &X, const std::vector< Index > &Xs, const Scope &A, const std::vector< Index > &As, const Scope &Y, const std::vector< Index > &Ys, const Scope &O, const std::vector< Index > &Os) const

const Scope &	GetYSoI_O (Index o) const

const Scope &	GetYSoI_Y (Index y) const

Index	IndividualToJointActionIndices (const Index *AI_ar) const
	Returns the joint action index that corresponds to the array of specified individual action indices. More...

Index	IndividualToJointActionIndices (const std::vector< Index > &indivActionIndices) const
	Returns the joint action index that corresponds to the vector of specified individual action indices. More...

Index	IndividualToJointActionIndices (const std::vector< Index > &ja_e, const Scope &agSC) const
	indiv->joint for a restricted set (Scope) of agents More...

Index	IndividualToJointObservationIndices (const std::vector< Index > &indivObservationIndices) const
	Returns the joint observation index that corresponds to the vector of specified individual observation indices. More...

Index	IndividualToJointObservationIndices (const std::vector< Index > &jo_e, const Scope &agSC) const
	indiv->joint for a restricted set (Scope) of agents More...

virtual void	Initialize2DBN ()

virtual void	Initialize2DBN (ScopeFunctor &SetScopes, TransitionProbFunctor &ComputeTransitionProb, ObservationProbFunctor &ComputeObservationProb)
	This signature allows us to initialize the 2DBN using externally supplied functors to set the scopes, and compute transition and observation probabilities in a discrete factored model. More...

bool	JointAIndicesValid () const

bool	JointIndicesValid () const

bool	JointOIndicesValid () const

const std::vector< Index > &	JointToIndividualActionIndices (Index jaI) const
	Returns a vector of indices to indiv. More...

std::vector< Index >	JointToIndividualActionIndices (Index ja_e, const Scope &agSC) const
	joint->indiv for a restricted set (Scope) of agents More...

const std::vector< Index > &	JointToIndividualObservationIndices (Index joI) const
	Returns a vector of indices to indiv. More...

std::vector< Index >	JointToIndividualObservationIndices (Index jo_e, const Scope &agSC) const
	joint->indiv for a restricted set (Scope) of agents More...

Index	JointToRestrictedJointActionIndex (Index jaI, const Scope &agSc_e) const
	Converts a global joint action index jaI to a restricted joint action index ja_e, for agents scope agSc_e Returns a vector of indices to indiv. More...

Index	JointToRestrictedJointObservationIndex (Index joI, const Scope &agSc_e) const
	Converts a global joint observation index joI to a restricted joint observation index jo_e, for agents scope agSc_e Returns a vector of indices to indiv. More...

void	MarginalizeTransitionObservationModel (const Index sf, bool sparse)
	This function marginalizes a state factor out of the flat joint transition and observation models of the system. More...

	MultiAgentDecisionProcessDiscreteFactoredStates (const std::string &name="received unspec. by MultiAgentDecisionProcessDiscreteFactoredStates", const std::string &descr="received unspec. by MultiAgentDecisionProcessDiscreteFactoredStates", const std::string &pf="received unspec. by MultiAgentDecisionProcessDiscreteFactoredStates")
	(default) Constructor More...

	MultiAgentDecisionProcessDiscreteFactoredStates (const MultiAgentDecisionProcessDiscreteFactoredStates &a)
	Copy constructor. More...

MultiAgentDecisionProcessDiscreteFactoredStates &	operator= (const MultiAgentDecisionProcessDiscreteFactoredStates &o)
	Copy assignment operator. More...

void	RemoveStateFactor (Index sf)
	This function removes a state factor from the model's MADPComponentFactoredStates, fixes the factor indices, and adjusts the 2DBN accordingly - all CPTs of nodes which depend on the removed state factor are recalculated by marginalizing their respective clique joints. More...

Index	SampleInitialState () const
	Sample a state according to the initial state PDF. More...

void	SampleInitialState (std::vector< Index > &sIs) const

Index	SampleJointObservation (Index jaI, Index sucI) const
	Sample an observation - needed for simulations. More...

Index	SampleJointObservation (Index sI, Index jaI, Index sucI) const

void	SampleJointObservation (const std::vector< Index > &aIs, const std::vector< Index > &sucIs, std::vector< Index > &oIs) const

void	SampleJointObservation (const std::vector< Index > &sIs, const std::vector< Index > &aIs, const std::vector< Index > &sucIs, std::vector< Index > &oIs) const

Index	SampleSuccessorState (Index sI, Index jaI) const
	Sample a successor state - needed by simulations. More...

void	SampleSuccessorState (const std::vector< Index > &sIs, const std::vector< Index > &aIs, std::vector< Index > &sucIs) const

bool	SetActionsInitialized (bool b)

void	SetConnectionsSpecified (bool b)

void	SetEventObservability (bool eventO)

bool	SetInitialized (bool b)

void	SetISD (FactoredStateDistribution *p)

void	SetNrActions (Index AI, size_t nrA)

void	SetNrObservations (Index AI, size_t nrO)

bool	SetObservationsInitialized (bool b)

void	SetSoI_O (Index o, const Scope &ASoI, const Scope &YSoI, const Scope &OSoI)

void	SetSoI_O (Index o, const Scope &XSoI, const Scope &ASoI, const Scope &YSoI, const Scope &OSoI)

void	SetSoI_Y (Index y, const Scope &XSoI, const Scope &ASoI, const Scope &YSoI)

void	SetSparse (bool b)

bool	SetStatesInitialized (bool b)

void	SetUniformISD ()

std::string	SoftPrint () const
	SoftPrints information on the MultiAgentDecisionProcessDiscrete. More...

std::string	SoftPrintState (Index sI) const

Index	StateIndexToFactorValueIndex (Index factor, Index s) const
	Get the value of a particular state factor given a joint flat state. More...

std::vector< Index >	StateIndexToFactorValueIndices (Index stateI) const
	Get the vector of FactorValue indices corresponding to stateI used to be called virtual vector<Index> GetStateFactorValues(Index stateI) const. More...

std::vector< Index >	StateIndexToFactorValueIndices (Index s_e, const Scope &sfSC) const
	convert an local state index s_e to a vector of state factors (with scope sfScope). More...

Scope	StateScopeBackup (const Scope &stateScope, const Scope &agentScope) const

	~MultiAgentDecisionProcessDiscreteFactoredStates ()
	Destructor. More...

Public Member Functions inherited from MultiAgentDecisionProcess
void	AddAgent (std::string name)
	Add a new agent with name "name" to the MADP. More...

Index	GetAgentIndexByName (const std::string &s) const
	Returns the index of an agent given its name, if it exists. More...

std::string	GetAgentNameByIndex (Index i) const
	Returns the name of the agent at the given index. More...

const Scope &	GetAllAgentScope () const

std::string	GetDescription () const

std::string	GetName () const

size_t	GetNrAgents () const
	Returns the number of agents in this MultiAgentDecisionProcess. More...

std::string	GetProblemFile () const
	Returns the name of the problem file. More...

std::string	GetUnixName () const
	Returns a name which can be in unix path, by default the base part of the problem filename. More...

	MultiAgentDecisionProcess (size_t nrAgents, const std::string &name="received unspec. by MultiAgentDecisionProcess", const std::string &description="received unspec. by MultiAgentDecisionProcess", const std::string &pf="received unspec. by MultiAgentDecisionProcess")
	Constructor. More...

	MultiAgentDecisionProcess (const std::string &name="received unspec. by MultiAgentDecisionProcess", const std::string &description="received unspec. by MultiAgentDecisionProcess", const std::string &pf="received unspec. by MultiAgentDecisionProcess")
	Default Constructor without specifying the number of agents. More...

void	Print () const

void	SetDescription (const std::string &description)

void	SetName (const std::string &name)

void	SetNrAgents (size_t nrAgents)
	Sets the number of agents. More...

void	SetUnixName (std::string unixName)
	Sets the name which can be used inin unix paths. More...

std::string	SoftPrint () const
	Prints some information on the MultiAgentDecisionProcess. More...

virtual	~MultiAgentDecisionProcess ()
	Destructor. More...

Public Member Functions inherited from DecPOMDP
	DecPOMDP ()
	Default constructor. sets RewardType to REWARD and discount to 1.0. More...

double	GetDiscount () const
	Returns the discount parameter. More...

double	GetDiscountForAgent (Index agentI) const
	Returns the discount parameter. More...

reward_t	GetRewardType () const
	Returns the reward type. More...

reward_t	GetRewardTypeForAgent (Index agentI) const
	Returns the reward type. More...

void	SetDiscount (double d)
	Sets the discount parameter to d. More...

void	SetDiscountForAgent (Index agentI, double d)
	Functions needed for POSGInterface: More...

void	SetRewardType (reward_t r)
	Sets the reward type to reward_t r. More...

void	SetRewardTypeForAgent (Index agentI, reward_t r)
	Sets the reward type to reward_t r. More...

std::string	SoftPrint () const
	SoftPrints some information on the DecPOMDP. More...

Static Protected Member Functions
static bool	ConsistentVectorsOnSpecifiedScopes (const std::vector< Index > &v1, const Scope &scope1, const std::vector< Index > &v2, const Scope &scope2)
	Auxiliary function that returns whether v1 is consistent with v2. More...

Private Attributes
std::vector< Scope >	_m_agScopes

bool	_m_cached_FlatRM

FactoredQFunctionScopeForStage	_m_immRewScope
	GetImmediateRewardScopes returns a reference to the following: More...

std::vector< RewardModel * >	_m_LRFs

std::vector< std::vector < size_t > >	_m_nrActionVals
	For each LRF, we maintain the nr of actions for each agent in its scope. More...

std::vector< size_t >	_m_nrAIs
	for each LRF we maintain the nr of action instantiations More...

size_t	_m_nrLRFs

std::vector< std::vector < size_t > >	_m_nrSFVals
	We maintain the nr of values for each state factor in the scope of lrf. More...

std::vector< size_t >	_m_nrXIs
	for each LRF we maintain the nr of X instantiations More...

RewardModel *	_m_p_rModel
	Pointer to model. More...

std::vector< Scope >	_m_sfScopes

bool	_m_sparse_FlatRM

Additional Inherited Members
Protected Member Functions inherited from MultiAgentDecisionProcessDiscreteFactoredStates
TwoStageDynamicBayesianNetwork *	Get2DBN ()

virtual bool	SanityCheckObservations () const

virtual bool	SanityCheckTransitions () const

Protected Attributes inherited from MultiAgentDecisionProcess
std::vector< Agent >	_m_agents
	Vector containing Agent objects, which are indexed named entities. More...

Scope	_m_allAgentsScope
	Scope containing all agents - usefull sometimes. More...

size_t	_m_nrAgents
	The number of agents participating in the MADP. More...

std::string	_m_problemFile
	String holding the filename of the problem file to be parsed - if applicable. More...

std::string	_m_unixName
	String for the unix name of the MADP. More...

Detailed Description

FactoredDecPOMDPDiscrete is implements a factored DecPOMDPDiscrete.

This class implements FactoredDecPOMDPDiscreteInterface is the interface for a Dec-POMDP with factored states. It defines the get/set reward functions.

This implementation maintains a vector of Local reward functions (LRFs), each of which is a RewardModel. The implementation of each RewardModel (which subclass of RewardModel is actually used) can differ per LRF.

Each reward model is defined over a subset of state factors and states: the state factor and agent scope of the particular LRF. This class is responsible of all index conversions. (so the referred RewardModels do not know anything about scopes etc.)

The `edges' in this code refer to edges of the graph constructed from the factored immediate reward function. (i.e., each `edge' corresponds to a local reward function)

Constructor & Destructor Documentation

FactoredDecPOMDPDiscrete::FactoredDecPOMDPDiscrete	(	std::string	name = `"received unspec. by FactoredDecPOMDPDiscrete"`,
		std::string	descr = `"received unspec. by FactoredDecPOMDPDiscrete"`,
		std::string	pf = `"received unspec. by FactoredDecPOMDPDiscrete"`
	)

Default constructor.

Constructor that sets the name, description, and problem file, and subsequently loads this problem file.

Referenced by Clone().

FactoredDecPOMDPDiscrete::~FactoredDecPOMDPDiscrete ( )

virtual

Destructor.

References _m_LRFs, and _m_p_rModel.

Member Function Documentation

void FactoredDecPOMDPDiscrete::CacheFlatModels ( bool sparse )

virtual

Reimplemented from MultiAgentDecisionProcessDiscreteFactoredStatesInterface.

Reimplemented in FactoredMMDPDiscrete.

References MultiAgentDecisionProcessDiscreteFactoredStates::CacheFlatObservationModel(), CacheFlatRewardModel(), MultiAgentDecisionProcessDiscreteFactoredStates::CacheFlatTransitionModel(), MultiAgentDecisionProcessDiscreteFactoredStates::ConstructJointActions(), and MultiAgentDecisionProcessDiscreteFactoredStates::ConstructJointObservations().

Referenced by ArgumentUtils::GetDecPOMDPDiscreteInterfaceFromArgs(), and ArgumentUtils::GetFactoredDecPOMDPDiscreteInterfaceFromArgs().

void FactoredDecPOMDPDiscrete::CacheFlatRewardModel ( bool sparse = false )

References _m_cached_FlatRM, _m_p_rModel, _m_sparse_FlatRM, MultiAgentDecisionProcessDiscreteFactoredStates::GetNrJointActions(), MultiAgentDecisionProcessDiscreteFactoredStates::GetNrStates(), GetReward(), Globals::REWARD_PRECISION, and RewardModelDiscreteInterface::Set().

Referenced by CacheFlatModels().

void FactoredDecPOMDPDiscrete::ClipRewardModel	(	Index	sf,
		bool	sparse
	)

Adjusts the reward model by ignoring a state factor.

References _m_cached_FlatRM, _m_p_rModel, _m_sparse_FlatRM, MultiAgentDecisionProcessDiscreteFactoredStates::GetNrJointActions(), MultiAgentDecisionProcessDiscreteFactoredStates::GetNrStateFactors(), MultiAgentDecisionProcessDiscreteFactoredStates::GetNrValuesForFactor(), GetReward(), Globals::REWARD_PRECISION, and RewardModelDiscreteInterface::Set().

Referenced by MarginalizeStateFactor().

virtual FactoredDecPOMDPDiscrete* FactoredDecPOMDPDiscrete::Clone ( ) const

inlinevirtual

Returns a pointer to a copy of this class.

Implements FactoredDecPOMDPDiscreteInterface.

Reimplemented in ProblemFireFightingFactored, ProblemAloha, ProblemFOBSFireFightingFactored, and FactoredMMDPDiscrete.

References FactoredDecPOMDPDiscrete().

virtual double FactoredDecPOMDPDiscrete::ComputeObservationProb	(	Index	o,
		Index	oVal,
		const std::vector< Index > &	As,
		const std::vector< Index > &	Ys,
		const std::vector< Index > &	Os
	)		const

inlinevirtual

Implements MultiAgentDecisionProcessDiscreteFactoredStates.

Reimplemented in ProblemAloha, ProblemFireFightingFactored, and FactoredMMDPDiscrete.

virtual double FactoredDecPOMDPDiscrete::ComputeTransitionProb	(	Index	y,
		Index	yVal,
		const std::vector< Index > &	Xs,
		const std::vector< Index > &	As,
		const std::vector< Index > &	Ys
	)		const

inlinevirtual

Implements MultiAgentDecisionProcessDiscreteFactoredStates.

Reimplemented in ProblemAloha, ProblemFireFightingFactored, and ProblemFOBSFireFightingFactored.

Referenced by FactoredMMDPDiscrete::Initialize2DBN().

bool FactoredDecPOMDPDiscrete::ConsistentVectorsOnSpecifiedScopes	(	const std::vector< Index > &	v1,
		const Scope &	scope1,
		const std::vector< Index > &	v2,
		const Scope &	scope2
	)

staticprotected

Auxiliary function that returns whether v1 is consistent with v2.

The scope of v2 (scope 2) should be a subset of scope 1

References Scope::GetPositionForIndex().

Referenced by SetRewardForLRF().

void FactoredDecPOMDPDiscrete::ConvertFiniteToInfiniteHorizon ( size_t horizon )

inline

void FactoredDecPOMDPDiscrete::CreateNewRewardModel ( )

inlinevirtual

Creates a new reward model mapping.

Implements DecPOMDPDiscreteInterface.

Referenced by CreateNewRewardModelForAgent().

void FactoredDecPOMDPDiscrete::CreateNewRewardModelForAgent ( Globals::Index )

inline

References CreateNewRewardModel().

void FactoredDecPOMDPDiscrete::ExportSpuddFile ( const std::string & filename ) const

virtual

Export fully-observable subset of this FactoredDecPOMDPDiscrete to spudd file.

This function is virtual since some specific problem instances may want to override variable names or other elements to conform with the spudd file format.

const Scope& FactoredDecPOMDPDiscrete::GetAgentScopeForLRF ( Index LRF ) const

inlinevirtual

Get all the agents in the scope of an LRF.

Implements FactoredDecPOMDPDiscreteInterface.

Referenced by ExportSpuddFile(), GetLRFRewardFlat(), ProblemAloha::InitializeAloha(), InitializeInstantiationInformation(), ProblemFOBSFireFightingFactored::InitializePFFF(), and ProblemFireFightingFactored::InitializePFFF().

const FactoredQFunctionScopeForStage FactoredDecPOMDPDiscrete::GetImmediateRewardScopes ( ) const

virtual

Returns all scopes of the immediate reward function in one object.

Implements FactoredDecPOMDPDiscreteInterface.

References _m_agScopes, _m_nrLRFs, _m_sfScopes, and FactoredQFunctionScopeForStage::AddLocalQ().

RewardModel* FactoredDecPOMDPDiscrete::GetLRF ( Index LRF ) const

inlinevirtual

Returns a pointer to the LRF-th reward model component.

Implements FactoredDecPOMDPDiscreteInterface.

double FactoredDecPOMDPDiscrete::GetLRFReward	(	Index	LRF,
		Index	s_e,
		Index	ja_e
	)		const

inlinevirtual

Returns reward for LRF, given ja_e - the joint index for a group action for the subset of agents as specified by the agent scope of LRF.

s_e - the (joint) index of the subset of factors specified by the state factor scope of LRF.

For instance, let the agents scope of LRF be <1,3>, then group action <3,5> means that agent 1 select action 3, while agent 3 performs its 5th action. Using indextools we can find agSc_jaI. E.g. agSc_jaI = IndividualToJointIndices( <3,5>, <6,6> ) (where <6,6> is a vector which specifies the number of actions per agent, see IndexTools.h for more info).

Implements FactoredDecPOMDPDiscreteInterface.

Referenced by ExportSpuddFile(), GetLRFReward(), GetLRFRewardFlat(), and SetRewardForLRF().

double FactoredDecPOMDPDiscrete::GetLRFReward	(	Index	LRF,
		const std::vector< Index > &	sI_e,
		const std::vector< Index > &	jaI_e
	)		const

virtual

Implements FactoredDecPOMDPDiscreteInterface.

References GetLRFReward(), RestrictedActionVectorToJointIndex(), and RestrictedStateVectorToJointIndex().

double FactoredDecPOMDPDiscrete::GetLRFRewardFlat	(	Index	LRF,
		Index	flat_s,
		Index	full_ja
	)		const

virtual

might be necessary? Returns reward for LRF, given a flat state index, and a full joint action.

Implements FactoredDecPOMDPDiscreteInterface.

References MultiAgentDecisionProcessDiscreteFactoredStates::JointToIndividualActionIndices(), and MultiAgentDecisionProcessDiscreteFactoredStates::StateIndexToFactorValueIndices().

Referenced by GetReward().

double FactoredDecPOMDPDiscrete::GetLRFRewardFlat	(	Index	lrf,
		const std::vector< Index > &	sfacs,
		const std::vector< Index > &	as
	)		const

References GetAgentScopeForLRF(), GetLRFReward(), GetStateFactorScopeForLRF(), and IndexTools::RestrictIndividualIndicesToScope().

size_t FactoredDecPOMDPDiscrete::GetNrAIs ( Index LRF ) const

Get the number of action instantiations. I.

I.e., the number of local joint (group) actions Should be called after InitializeInstantiationInformation()

References _m_nrAIs.

Referenced by ProblemAloha::InitializeAloha(), ProblemFOBSFireFightingFactored::InitializePFFF(), and ProblemFireFightingFactored::InitializePFFF().

size_t FactoredDecPOMDPDiscrete::GetNrLRFs ( ) const

inlinevirtual

Get the number of LRFs in the factored representation.

Implements FactoredDecPOMDPDiscreteInterface.

References _m_nrLRFs.

Referenced by ExportSpuddFile(), GetReward(), and InitializeInstantiationInformation().

size_t FactoredDecPOMDPDiscrete::GetNrXIs ( Index LRF ) const

Get the number of X instantiations for LRF.

Each LRF has a local state space X. This function returns the size of this local state space. Should be called after InitializeInstantiationInformation()

References _m_nrXIs.

Referenced by ProblemAloha::InitializeAloha(), ProblemFOBSFireFightingFactored::InitializePFFF(), and ProblemFireFightingFactored::InitializePFFF().

double FactoredDecPOMDPDiscrete::GetReward	(	const std::vector< Index > &	sIs,
		const std::vector< Index > &	aIs
	)		const

virtual

Implements FactoredDecPOMDPDiscreteInterface.

References GetLRFRewardFlat(), and GetNrLRFs().

Referenced by CacheFlatRewardModel(), ClipRewardModel(), GetReward(), GetRewardForAgent(), and SetReward().

double FactoredDecPOMDPDiscrete::GetReward	(	Index	sI,
		Index	jaI
	)		const

virtual

Return the reward for state, joint action indices.

Implements DecPOMDPDiscreteInterface.

References _m_cached_FlatRM, _m_p_rModel, RewardModelDiscreteInterface::Get(), GetLRFRewardFlat(), and GetNrLRFs().

double FactoredDecPOMDPDiscrete::GetReward	(	State *	s,
		JointAction *	ja
	)		const

inlinevirtual

Function that returns the reward for a state and joint action.

This should be very generic.

Implements DecPOMDPInterface.

References GetReward().

double FactoredDecPOMDPDiscrete::GetRewardForAgent	(	Index	agentI,
		Index	sI,
		Index	jaI
	)		const

inline

Return the reward for state, joint action indices.

References GetReward().

double FactoredDecPOMDPDiscrete::GetRewardForAgent	(	Index	agentI,
		State *	s,
		JointAction *	ja
	)		const

inlinevirtual

Function that returns the reward for a state and joint action.

This should be very generic.

Implements POSGInterface.

References GetReward().

RGet * FactoredDecPOMDPDiscrete::GetRGet ( ) const

virtual

Implements DecPOMDPDiscreteInterface.

References _m_cached_FlatRM, _m_p_rModel, and _m_sparse_FlatRM.

const Scope& FactoredDecPOMDPDiscrete::GetStateFactorScopeForLRF ( Index LRF ) const

inlinevirtual

Get all the state factors in the scope of an LRF.

Implements FactoredDecPOMDPDiscreteInterface.

Referenced by ExportSpuddFile(), GetLRFRewardFlat(), ProblemAloha::InitializeAloha(), InitializeInstantiationInformation(), ProblemFOBSFireFightingFactored::InitializePFFF(), and ProblemFireFightingFactored::InitializePFFF().

void FactoredDecPOMDPDiscrete::InitializeInstantiationInformation ( )

Initializes some meta information computed from the scopes.

This function computes the number of X and A instantiations, the vectors with the number of values for sfacs and step-size arrays used in conversion from individual <-> joint indices.

References _m_nrActionVals, _m_nrAIs, _m_nrSFVals, _m_nrXIs, GetAgentScopeForLRF(), MultiAgentDecisionProcessDiscreteFactoredStates::GetNrActions(), GetNrLRFs(), MultiAgentDecisionProcessDiscreteFactoredStates::GetNrValuesPerFactor(), GetStateFactorScopeForLRF(), and IndexTools::RestrictIndividualIndicesToScope().

Referenced by ProblemAloha::InitializeAloha(), ProblemFOBSFireFightingFactored::InitializePFFF(), and ProblemFireFightingFactored::InitializePFFF().

void FactoredDecPOMDPDiscrete::InitializeStorage ( )

makes sure that _m_sfScopes, _m_agScopes, _m_LRFs are set to proper size

References _m_agScopes, _m_LRFs, _m_nrActionVals, _m_nrAIs, _m_nrLRFs, _m_nrSFVals, _m_nrXIs, and _m_sfScopes.

Referenced by SetNrLRFs().

void FactoredDecPOMDPDiscrete::MarginalizeISD	(	Index	sf,
		std::vector< size_t > &	factor_sizes,
		const FactoredStateDistribution *	fsd
	)

Marginalizes the ISD over a given state factor.

References MultiAgentDecisionProcessDiscreteFactoredStates::GetISD(), FactoredStateDistribution::GetProbability(), IndexTools::Increment(), Globals::PROB_PRECISION, FSDist_COF::SanityCheck(), MultiAgentDecisionProcessDiscreteFactoredStates::SetISD(), and FSDist_COF::SetProbability().

Referenced by MarginalizeStateFactor().

void FactoredDecPOMDPDiscrete::MarginalizeStateFactor	(	Index	sf,
		bool	sparse
	)

Attempts to remove an extraneous state factor from the problem.

Such state factors are typically useful when modeling large systems graphically, but hinder performance when solving with flat algorithms (such as Perseus). Currently, it only supports the marginalization of nodes without NS dependencies, and which do not directly influence any LRF

References ClipRewardModel(), MultiAgentDecisionProcessDiscreteFactoredStates::GetFactoredISD(), MultiAgentDecisionProcessDiscreteFactoredStates::GetNrValuesPerFactor(), MarginalizeISD(), and MultiAgentDecisionProcessDiscreteFactoredStates::MarginalizeTransitionObservationModel().

Referenced by ArgumentUtils::GetFactoredDecPOMDPDiscreteInterfaceFromArgs().

void FactoredDecPOMDPDiscrete::Print ( ) const

inlinevirtual

Implements FactoredDecPOMDPDiscreteInterface.

References SoftPrint().

Index FactoredDecPOMDPDiscrete::RestrictedActionVectorToJointIndex	(	Index	LRF,
		const std::vector< Index > &	actionVec_e
	)		const

virtual

convert an action vector of restricted scope to a joint index a_e.

This is a convenience function that performs indiv->joint action index conversion for a specific edge e (LRF). (i.e., this function is typically called when requesting the immediate reward)

actionVec_e is an assignment of all actions in the agent scope scope of e.
the funtion returns a joint (group) index a_e.

Implements FactoredDecPOMDPDiscreteInterface.

References _m_nrActionVals, and IndexTools::IndividualToJointIndices().

Referenced by GetLRFReward(), and SetRewardForLRF().

Index FactoredDecPOMDPDiscrete::RestrictedStateVectorToJointIndex	(	Index	LRF,
		const std::vector< Index > &	stateVec_e
	)		const

virtual

Get the vector of Factor indices corresponding to stateI.

convert a state vector of restricted scope to a joint index s_e.

This is a convenience function that performs indiv->joint state index conversion for a specific edge e (LRF).

stateVec_e is an assignment of all state factors in the state factor scope of e.
the funtion returns a joint (group) index s_e.

Implements FactoredDecPOMDPDiscreteInterface.

References _m_nrSFVals, and IndexTools::IndividualToJointIndices().

Referenced by GetLRFReward(), and SetRewardForLRF().

void FactoredDecPOMDPDiscrete::SetNrLRFs ( size_t nr )

inline

References InitializeStorage().

Referenced by ProblemAloha::InitializeAloha(), ProblemFOBSFireFightingFactored::InitializePFFF(), and ProblemFireFightingFactored::InitializePFFF().

virtual void FactoredDecPOMDPDiscrete::SetOScopes ( )

inlinevirtual

Implements MultiAgentDecisionProcessDiscreteFactoredStates.

Reimplemented in ProblemFireFightingFactored, ProblemAloha, FactoredMMDPDiscrete, and ProblemFireFightingGraph.

void FactoredDecPOMDPDiscrete::SetReward	(	Index	sI,
		Index	jaI,
		double	r
	)

inlinevirtual

Set the reward for state, joint action indices.

Implements DecPOMDPDiscreteInterface.

Referenced by SetReward(), and SetRewardForAgent().

void FactoredDecPOMDPDiscrete::SetReward	(	Index	sI,
		Index	jaI,
		Index	sucSI,
		double	r
	)

virtual

Set the reward for state, joint action , suc. state indices.

Implements DecPOMDPDiscreteInterface.

References GetReward(), MultiAgentDecisionProcessDiscreteFactoredStates::GetTransitionProbability(), and SetReward().

void FactoredDecPOMDPDiscrete::SetReward	(	Index	sI,
		Index	jaI,
		Index	sucSI,
		Index	joI,
		double	r
	)

virtual

Set the reward for state, joint action, suc.state, joint obs indices.

Implements DecPOMDPDiscreteInterface.

void FactoredDecPOMDPDiscrete::SetReward	(	State *	s,
		JointAction *	ja,
		double	r
	)

inlinevirtual

Function that sets the reward for a state and joint action.

This should be very generic.

Implements DecPOMDPInterface.

References SetReward().

void FactoredDecPOMDPDiscrete::SetRewardForAgent	(	Index	agentI,
		Index	sI,
		Index	jaI,
		double	r
	)

inline

References SetReward().

void FactoredDecPOMDPDiscrete::SetRewardForAgent	(	Index	agentI,
		Index	sI,
		Index	jaI,
		Index	sucSI,
		double	r
	)

inline

Set the reward for state, joint action , suc. state indices.

References SetReward().

void FactoredDecPOMDPDiscrete::SetRewardForAgent	(	Index	agentI,
		Index	sI,
		Index	jaI,
		Index	sucSI,
		Index	joI,
		double	r
	)

inline

Set the reward for state, joint action, suc.state, joint obs indices.

References SetReward().

void FactoredDecPOMDPDiscrete::SetRewardForAgent	(	Index	agentI,
		State *	s,
		JointAction *	ja,
		double	r
	)

inlinevirtual

Function that sets the reward for an agent, state and joint action.

This should be very generic.

Implements POSGInterface.

References SetReward().

void FactoredDecPOMDPDiscrete::SetRewardForLRF	(	Index	LRF,
		const std::vector< Index > &	Xs,
		const std::vector< Index > &	As,
		double	reward
	)

Set reward.

The reward model should already be added. This function does not require any back-projection or whatsoever.

References _m_agScopes, _m_LRFs, _m_sfScopes, RestrictedActionVectorToJointIndex(), RestrictedStateVectorToJointIndex(), and PrintTools::SoftPrintVector().

Referenced by ProblemAloha::InitializeAloha(), ProblemFOBSFireFightingFactored::InitializePFFF(), ProblemFireFightingFactored::InitializePFFF(), and SetRewardForLRF().

void FactoredDecPOMDPDiscrete::SetRewardForLRF	(	Index	LRF,
		const std::vector< Index > &	Xs,
		const std::vector< Index > &	As,
		const Scope &	Y,
		const std::vector< Index > &	Ys,
		const Scope &	O,
		const std::vector< Index > &	Os,
		double	reward
	)

Set reward.

The reward model should already be added. Because we maintain reward functions of the form R(s,a), this function distributes the reward according the the expectation of y, o. I.e., Let X, A be the registered scope for LRF, and x,a be instantiations of those scopes. Then

$R(x,a) = \sum_{y,o} R(x,a,y,o) \Pr(y,o|x,a)$

Therefore this function performs

$R(x,a) += R(x,a,y,o) \Pr(y,o|x,a)$

(by calling SetRewardForLRF(Index LRF, const std::vector<Index>& Xs,const std::vector<Index>& As, double reward) )

Repeatedly calling this function (for all y, o) should result in a valid R(x,a).

This means that

Xs should be a vector of SFValue indices for registered scope _m_sfScopes[LRF].
As should be a vector specifying an action for each agent specified by agent scope _m_agScopes[LRF].
$\Pr(y,o|x,a)$ should be defined. (i.e., the scope backup of o and y should not involve variables not in X,A).

References _m_agScopes, _m_sfScopes, GetLRFReward(), MultiAgentDecisionProcessDiscreteFactoredStates::GetYOProbability(), and SetRewardForLRF().

void FactoredDecPOMDPDiscrete::SetRewardForLRF	(	Index	LRF,
		const Scope &	X,
		const std::vector< Index > &	Xs,
		const Scope &	A,
		const std::vector< Index > &	As,
		const Scope &	Y,
		const std::vector< Index > &	Ys,
		const Scope &	O,
		const std::vector< Index > &	Os,
		double	reward
	)

Set Reward.

This set reward function is best illustrated by an example of the factored firefighting problem [Oliehoek08AAMAS]. There we have that the reward function for the first house can be expressed as.

$\forall_{f_1,f_2,a_1} \quad R(f_1,f_2,a_1) = \sum_{f_1'} \Pr(f_1'|f_1,f_2,a_1) R(f_1')$

That is, if the scope is underspecified (X and A only are subsets of the registered scopes) than it holds {for all} non-specified factors and actions.

This version of SetRewardForLRF should be called with

X a subset of the registered state factor scope for LRF
A a subset of the registered agent scope for LRF

Consequently it will call void SetRewardForLRF(Index LRF, const vector<Index>& Xs, const vector<Index>& As, const Scope& Y, const vector<Index>& Ys, const Scope& O, const vector<Index>& Os, double reward ) for all non specified state factors and actions.

References _m_agScopes, _m_nrActionVals, _m_nrSFVals, _m_sfScopes, ConsistentVectorsOnSpecifiedScopes(), IndexTools::Increment(), and SetRewardForLRF().