Feature Selection Toolbox FST4
UTIA · Czech Academy of Sciences
Loading...
Searching...
No Matches
Public Types | Public Member Functions | Protected Attributes | List of all members
FST::Search_Monte_Carlo< RETURNTYPE, DIMTYPE, SUBSET, CRITERION > Class Template Reference

Repeatedly samples random subsets to eventually yield the one with highest criterion value. More...

#include <search_monte_carlo.hpp>

Inheritance diagram for FST::Search_Monte_Carlo< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >:
FST::Search< RETURNTYPE, DIMTYPE, SUBSET, CRITERION > FST::Search_Monte_Carlo_Threaded< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >

Public Types

typedef Search< RETURNTYPE, DIMTYPE, SUBSET, CRITERION > parent
 base class alias
typedef std::shared_ptr< CRITERION > PCriterion
 shared pointer to the criterion the search optimizes
typedef std::shared_ptr< SUBSET > PSubset
 shared pointer to a feature subset
typedef Candidate_Evaluator< RETURNTYPE, DIMTYPE, SUBSET, CRITERION > EVALUATORENGINE
 the injected evaluation-strategy interface (sequential in-line or thread-pooled)
typedef std::shared_ptr< EVALUATORENGINE > PEvaluatorEngine
Public Types inherited from FST::Search< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >
 shared pointer to the criterion the search optimizes
 shared pointer to a feature subset
 shared pointer to the result tracker enabled via enable_result_tracking()
 the observer interface type this search reports through (see Search_Observer)
 shared pointer to an OBSERVER, as set via set_observer()

Public Member Functions

 Search_Monte_Carlo (const unsigned long long trials_limit=1000, const unsigned long time_limit=0, const unsigned int time_limit_check_freq=1, PEvaluatorEngine engine=PEvaluatorEngine())
 engine defaults to sequential in-line evaluation if not supplied; see set_stopping_condition() for trials_limit / time_limit / time_limit_check_freq
virtual bool search (const DIMTYPE target_d, RETURNTYPE &result, const PSubset sub, const PCriterion crit, std::ostream &os=std::cout) override
 samples random subsets of target_d features (of randomized size if target_d==0: from the configured range or inclusion probability, see set_cardinality_randomization(), or uniformly from [1,n] when neither is set) until the stopping condition is met, and returns the best one + its criterion value; false if no candidate was evaluated successfully. target_d must be smaller than the full feature count - drawing all n features at random is not a search (unlike Search_Exhaustive, which accepts target_d==n)
void set_stopping_condition (const unsigned long long trials_limit=1000, const unsigned long time_limit=0, const unsigned int time_limit_check_freq=100)
 set maximum number of random trials or maximum time in seconds or both; trials_limit and time_limit both 0 is replaced by a limit of 1000 trials, time_limit_check_freq 0 by 100 (the elapsed time is read once per that many trials; the constructor asks for 1, i.e. a check per trial)
void set_cardinality_randomization (const DIMTYPE d_from, const DIMTYPE d_to)
 generate subset size first from [d_from,d_to], then randomly choose exactly that number of features
void set_cardinality_randomization (const float d_prob)
 for each feature roll the dice and select it with probability d_prob
void reset_cardinality_randomization ()
 clears both ways of randomizing the subset size (the constructor calls this); with neither of them in effect a target_d==0 search draws each probe's size uniformly from [1,n]
virtual std::ostream & print (std::ostream &os) const override
 writes a human-readable identification of the search and its configuration; base default writes nothing
Public Member Functions inherited from FST::Search< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >
 assigns a result tracker to the search process, i.e., enables logging of all evaluated subsets for future reuse (re-evaluation, etc.)
 detaches any result tracker (result_tracker_active() then returns false)
 true if a result tracker was assigned via enable_result_tracking() and not since disabled
 sets the amount of information that is logged to output stream throughout the course of search
 the currently configured output detail level
 true unless output detail is SILENT
 true only at the DETAILED output level
 redirects search progress reporting away from the default syncout text output to observer

Protected Attributes

PEvaluatorEngine _engine
 injected evaluation strategy (sequential-inline or thread-pooled); defaulted in the constructor if not supplied
unsigned long _time_limit
 max time allowed to be spent in seconds
unsigned long long _trials_limit
 max no of candidates evaluated (64-bit: counts candidates, see the LLP64 note on Candidate_Evaluator::BestCallback)
unsigned int _time_limit_check_freq
 number of trials before next time check
DIMTYPE _d_from
 lower bound of the random subset size range; see set_cardinality_randomization(const DIMTYPE, const DIMTYPE)
DIMTYPE _d_to
 upper bound of the random subset size range
float _d_prob
 active d_prob must be from (0,1]; generates random subset of random size where the mean size over a series of calls is d_prob*no_of_all_features
Protected Attributes inherited from FST::Search< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >
 the assigned result tracker, or null (see enable_result_tracking()/result_tracker_active())
 the output detail level set via set_output_detail(); read by output_normal()/output_detailed()
 the attached progress observer, or null (see set_observer(), report(), report_solution())

Additional Inherited Members

Protected Member Functions inherited from FST::Search< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >
 delivers one pre-formatted progress message: to the observer if attached, otherwise as syncout text to os
 like report(), for sites announcing a new (sub-)solution; fires on_solution instead

Detailed Description

template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class CRITERION>
class FST::Search_Monte_Carlo< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >

Repeatedly samples random subsets to eventually yield the one with highest criterion value.

The search always runs under a terminating condition - a number of random trials, a time limit in seconds, or both; with neither requested it stops after 1000 trials (see set_stopping_condition()). Candidate evaluation goes through the injected engine (a Candidate_Evaluator implementation, sequential in-line by default); the random draws themselves always happen on the calling thread, so the same seed yields the same sequence of candidates at any thread count.

Note
offers two ways of random subset generation: a) set_cardinality_randomization(const DIMTYPE d_from, const DIMTYPE d_to) invokes the first option, i.e., when generating a random subset, the subset size is first chosen randomly from [d_from,d_to], then features are randomly selected until the required number is reached b) set_cardinality_randomization(const float d_prob) invokes the second option, where each feature is randomly included/excluded with probability d_prob. The expected subset size is thus d_prob*no_of_all_features, but can be actually anything from [0,no_of_all_features].

Member Typedef Documentation

◆ PEvaluatorEngine

template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class CRITERION>
typedef std::shared_ptr<EVALUATORENGINE> FST::Search_Monte_Carlo< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >::PEvaluatorEngine

shared pointer to an EVALUATORENGINE

Member Function Documentation

◆ print()

template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class CRITERION>
std::ostream & FST::Search_Monte_Carlo< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >::print ( std::ostream & os) const
overridevirtual

writes a human-readable identification of the search and its configuration; base default writes nothing

Reimplemented from FST::Search< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >.

◆ search()

template<class RETURNTYPE, typename DIMTYPE, class SUBSET, class CRITERION>
bool FST::Search_Monte_Carlo< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >::search ( const DIMTYPE target_d,
RETURNTYPE & result,
const PSubset sub,
const PCriterion crit,
std::ostream & os = std::cout )
overridevirtual

samples random subsets of target_d features (of randomized size if target_d==0: from the configured range or inclusion probability, see set_cardinality_randomization(), or uniformly from [1,n] when neither is set) until the stopping condition is met, and returns the best one + its criterion value; false if no candidate was evaluated successfully. target_d must be smaller than the full feature count - drawing all n features at random is not a search (unlike Search_Exhaustive, which accepts target_d==n)

Implements FST::Search< RETURNTYPE, DIMTYPE, SUBSET, CRITERION >.

The documentation for this class was generated from the following file:
Generated by doxygen 1.17.0
Copyright © 2000–present ÚTIA, Institute of Information Theory and Automation of the Czech Academy of Sciences · Contacts · This site sets no cookies and uses no analytics.