This class represents the UCVE process. More...

#include <AIToolbox/Factored/Bandit/Algorithms/Utils/UCVE.hpp>

Classes
struct	Entry

Public Types
using	V = Eigen::Vector2d

using	Factor = std::vector< Entry >

using	Result = std::tuple< Action, V >

using	GVE = GenericVariableElimination< Factor >

Public Member Functions
template<typename Iterable >
Result	operator() (const Action &A, const double logtA, const Iterable &inputRules)
	This function is the entry point for the solving process. More...

Result	operator() (const Action &A, const double logtA, GVE::Graph &graph)
	This function performs the actual agent elimination process. More...

Detailed Description

This class represents the UCVE process.

This class implements variable elimination using bounds. It receives as input a series of rules, one per PartialAction, containing both an approximate value for the action, and a variance to keep track of how much the rule could be actually worth.

Internally, this uses a variation over MultiObjectiveVariableElimination, where the two objectives are the approximate value and the variance. Additionally, in order to be more efficient, the agent removal process during Variable Elimination computes an upper and lower variance bound for that agent, in order to remove actions which cannot possibly be optimal from the search.

As this class doesn't really do much cleanup after solving a process, it's probably best to create a separate instance for each solving process.

Member Typedef Documentation

◆ Factor

using AIToolbox::Factored::Bandit::UCVE::Factor = std::vector<Entry>

◆ GVE

using AIToolbox::Factored::Bandit::UCVE::GVE = GenericVariableElimination<Factor>

◆ Result

using AIToolbox::Factored::Bandit::UCVE::Result = std::tuple<Action, V>

◆ V

using AIToolbox::Factored::Bandit::UCVE::V = Eigen::Vector2d

Member Function Documentation

◆ operator()() [1/2]

template<typename Iterable >

Result AIToolbox::Factored::Bandit::UCVE::operator()	(	const Action &	A,
		const double	logtA,
		const Iterable &	inputRules
	)

inline

This function is the entry point for the solving process.

Template Parameters

Iterable An iterable object containing a series of Entry.

Parameters

A	The size of the action space
logtA	The precomputed value for logtA for this timestep.
inputRules	The rules on which the Variable Elimination process will work.

Returns: The optimal action to take, and its value.

◆ operator()() [2/2]

Result AIToolbox::Factored::Bandit::UCVE::operator()	(	const Action &	A,
		const double	logtA,
		GVE::Graph &	graph
	)

This function performs the actual agent elimination process.

Parameters

A	The size of the action space
logtA	The precomputed value for logtA for this timestep.
graph	The graph to perform UCVE on.

Returns: The best action, randomly taken if multiple actions are eligible.

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

include/AIToolbox/Factored/Bandit/Algorithms/Utils/UCVE.hpp

Classes

Public Types

Public Member Functions