freetensor::RandCtx< RNG > Class Template Reference

#include <rand_ctx.h>

Inheritance diagram for freetensor::RandCtx< RNG >:

Collaboration diagram for freetensor::RandCtx< RNG >:

Public Member Functions
	RandCtx (RNG &rng)
int	decide (ProgramPosition pos, const std::string &name, const RandCondStack &condStack, const std::vector< double > &priori, const Ref< RandTrace > &trace, const std::string &message="")
Public Member Functions inherited from freetensor::RandCtxImpl
void	observeTrace (const Ref< RandTrace > &trace, double value, double stddev)
void	setLearnFilter (const std::regex &toLearn)
void	setLearn ()
void	setInfer ()

Additional Inherited Members
Protected Attributes inherited from freetensor::RandCtxImpl
std::unordered_map< ProgramPosition, std::unordered_map< Ref< RandCondInterface >, Ref< DiscreteRandVar >, PtrInvocable< std::hash< RandCondInterface > >, PtrInvocable< std::equal_to< RandCondInterface > > > >	randVars_
std::unordered_map< ProgramPosition, Ref< std::vector< int > > >	totCnt_
std::multimap< Ref< RandTrace >, std::pair< double, double >, PtrInvocable< std::less< RandTrace > > >	traces_
bool	isInfer_ = true
std::regex	toLearn_ {".*"}
std::mutex	lock_

Detailed Description

template<std::uniform_random_bit_generator RNG>
class freetensor::RandCtx< RNG >

Context to do random decisions

If setLearned, sample a decision and use the decision traces to learn a Bayesian model. Each trace is labeled with a performance value, and traces are compared pairwisely (lower is better), to learn P(this decision leads to better performance).

Constructor & Destructor Documentation

RandCtx()

template<std::uniform_random_bit_generator RNG>

freetensor::RandCtx< RNG >::RandCtx ( RNG &  rng )

inline

Member Function Documentation

decide()

template<std::uniform_random_bit_generator RNG>

int freetensor::RandCtx< RNG >::decide ( ProgramPosition  pos, const std::string &  name, const RandCondStack &  condStack, const std::vector< double > &  priori, const Ref< RandTrace > &  trace, const std::string &  message = ""  )

inline

Get decision from a random variable uniquely defined by conditions and the program position

If learning, sample a random variable and record it as a trace. If infering, pick a most likely decision

This function is thread-safe

---

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

• include/probability/rand_ctx.h