demo11t.cpp File Reference

Example 11t: Threaded wrapper-based feature selection with Floating Search. More...

#include <boost/smart_ptr.hpp>
#include <exception>
#include <iostream>
#include <cstdlib>
#include <string>
#include <vector>
#include "error.hpp"
#include "global.hpp"
#include "subset.hpp"
#include "data_intervaller.hpp"
#include "data_splitter.hpp"
#include "data_splitter_5050.hpp"
#include "data_splitter_cv.hpp"
#include "data_scaler.hpp"
#include "data_scaler_void.hpp"
#include "data_accessor_splitting_memTRN.hpp"
#include "data_accessor_splitting_memARFF.hpp"
#include "criterion_wrapper.hpp"
#include "distance_euclid.hpp"
#include "classifier_knn.hpp"
#include "seq_step_straight_threaded.hpp"
#include "search_seq_sffs.hpp"

Include dependency graph for demo11t.cpp:

Functions
int	main ()

---

Detailed Description

Example 11t: Threaded wrapper-based feature selection with Floating Search.

---

Function Documentation

int main

(

)

Example 11t: Threaded wrapper-based feature selection with Floating Search

As alternative to standard sequential evaluation of feature subset candidates FST3 enables threaded candidate subset evaluation. All FST3 sequential search methods can be easily parallelized by using Sequential_Step_Straight_Threaded instead of Sequential_Step_Straight evaluator object. The actual search speed gain depends on particular problem setting. In small-sample, low-dimensional settings, or when criterion evaluation is very fast, the actual gain can remain negligible or even negative due to thread management overhead (note that this also considerably depends on compiler). This example is the threaded version of Example 11: Wrapper-based feature selection with Floating Search. with the 15-dimensional (i.e., low-dimensional) small-sample speech dataset. It illustrates well the threading performance issue. When compiled under cygwin with gcc 4.3.x and Boost 1.33.1 it operates slower than the equivalent sequential code in Example 11: Wrapper-based feature selection with Floating Search.. In contrary, when compiled under Visual C++ 2010 with Boost 1.44, it operates faster.

Note:

Threading is generally capable of substantially accelerating the search process when applied in computationally more demanding scenarios; see Example 12t: Threaded SVM-wrapper-based feature selection with Dynamic Oscillating Search. or Example 25t: Fast pre-selection followed by Dynamic Oscillating Search. or Example 33t: Threaded Oscillating Search in very high-dimensional feature selection. or Example 40t: Threaded Exhaustive (optimal) feature selection., etc., for such cases.

The maximum permitted number of threads to run at once is to be user-specified with respect to hardware capabilities.

References FST::Search_SFFS< RETURNTYPE, DIMTYPE, SUBSET, CRITERION, EVALUATOR >::get_result(), FST::Search_SFFS< RETURNTYPE, DIMTYPE, SUBSET, CRITERION, EVALUATOR >::search(), and FST::Search_SFFS< RETURNTYPE, DIMTYPE, SUBSET, CRITERION, EVALUATOR >::set_search_direction().