例として最も単純な分類器を使用した言語パフォーマンスの比較

import std.stdio; import std.math; import std.format; import std.datetime; //version=COLLECT_MEASURES; //version=MERGE_CLASSES; struct Measure { float x=0, y=0; auto opBinary(string op)( auto ref const Measure m ) const if( op == "+" || op == "-" ) { mixin( "return Measure( x " ~ op ~ " mx, y " ~ op ~ " my );" ); } auto opBinary(string op)( float m ) const if( op == "*" || op == "/" ) { mixin( "return Measure( x " ~ op ~ " m, y " ~ op ~ " m );" ); } } unittest { auto a = Measure(1,3); auto b = Measure(4,5); auto add = a + b; assert( add.x == 5 && add.y == 8 ); auto dif = a - b; assert( dif.x == -3 && dif.y == -2 ); auto mlt = a * 3; assert( mlt.x == 3 && mlt.y == 9 ); auto div = b / 2; assert( div.x == 2 && div.y == 2.5 ); } float sqr( float v ) { return v * v; } float dist( ref const Measure a, ref const Measure b ) { return sqrt( sqr(ax - bx) + sqr(ay - by) ); } class Class { Measure mean; size_t N; version(COLLECT_MEASURES) Measure[] measures; this( in Measure m ) { mean = m; N = 1; version(COLLECT_MEASURES) measures ~= m; } void append( in Measure m ) { mean = ( mean * N + m ) / ( N + 1 ); N++; version(COLLECT_MEASURES) measures ~= m; } void merge( const Class m ) { mean = ( mean * N + m.mean * mN ) / ( N + mN ); N += mN; version(COLLECT_MEASURES) measures ~= m.measures; } } unittest { auto cls = new Class( Measure(1,2) ); assert( cls.mean.x == 1 && cls.mean.y == 2 ); assert( cls.N == 1 ); cls.append( Measure(3,4) ); assert( cls.mean.x == 2 && cls.mean.y == 3 ); assert( cls.N == 2 ); } class Classifier { public: Class[] list; float ncls_dist; this( float mdist ) { ncls_dist = mdist; } void classificate( ref const Measure m ) { float min_dist = float.max; Class near_cls; foreach( i; list ) { float d = dist( m, i.mean ); if( d < min_dist ) { min_dist = d; near_cls = i; } } if( min_dist < ncls_dist ) near_cls.append(m); else list ~= new Class(m); } void mergeClasses() { Class[] uniq; l: foreach( cls; list ) { foreach( trg; uniq ) if( dist( cls.mean, trg.mean ) < ncls_dist ) { trg.merge( cls ); continue l; } uniq ~= cls; } list = uniq; } } Measure[] readMeasuresFromFile( string name ) { auto file = File( name, "r" ); Measure[] list; foreach( line; file.byLine() ) { Measure tmp; formattedRead( line, "%f %f", &tmp.x, &tmp.y ); list ~= tmp; } return list; } void main( string[] args ) { auto measures = readMeasuresFromFile( args[1] ); StopWatch sw; sw.start(); auto clsf = new Classifier(3); foreach( m; measures ) clsf.classificate(m); version(MERGE_CLASSES) clsf.mergeClasses(); sw.stop(); writeln( "work time: ", sw.peek.nsecs / 1_000_000_000.0f ); foreach( cls; clsf.list ) writefln( "[%f, %f]: %d", cls.mean.x, cls.mean.y, cls.N ); }
      
      

        
        
        
      

    
        
        
        
      
      

        
        
        
      

    
     

 #![feature(duration)] #![feature(duration_span)] #![feature(append)] use std::ops::*; use std::borrow::Borrow; use std::f32; use std::cell::RefCell; use std::fs::File; use std::io::BufReader; use std::io::BufRead; use std::time::Duration; use std::env; #[derive(Clone,Copy)] struct Measure { x : f32, y : f32 } impl Measure{ #[allow(non_snake_case)] pub fn new(X:f32,Y:f32) -> Measure{ Measure{ x:X, y:Y } } } impl Add for Measure{ type Output = Measure; fn add(self, rhs:Measure) -> Measure { Measure{ x: self.x + rhs.x, y: self.y + rhs.y, } } } impl Sub for Measure{ type Output = Measure; fn sub(self, rhs: Measure) -> Measure { Measure{ x: self.x - rhs.x, y: self.y - rhs.y } } } impl Div<f32> for Measure { type Output = Measure; fn div(self, rhs: f32) -> Measure { Measure{ x: self.x / rhs, y: self.y / rhs } } } impl Mul<f32> for Measure { type Output = Measure; fn mul(self, rhs: f32) -> Measure { Measure{ x: self.x * rhs, y: self.y * rhs } } } #[inline] fn sqr(v:f32)->f32 { v * v } fn dist (a: & Measure, b: & Measure) -> f32 { (sqr(ax - bx) + sqr(ay - by)).sqrt() } #[derive(Clone)] struct Class { mean: Measure, count: usize, #[cfg(collect_measures)] measures: Vec<Measure> } impl Class{ #[cfg(collect_measures)] pub fn new( m: Measure ) -> Class { Class{ mean: m, count: 1, measures: vec![m.clone()] } } #[cfg(not(collect_measures))] pub fn new( m: Measure ) -> Class { Class{ mean: m, count: 1, } } #[cfg(collect_measures)] pub fn append(&mut self, m: Measure){ self.mean = ( self.mean * self.count as f32 + m ) /( self.count + 1)as f32; self.measures.push(m.clone()); self.count += 1; } #[cfg(not(collect_measures))] pub fn append(&mut self, m: Measure){ self.mean = ( self.mean * self.count as f32 + m ) /( self.count + 1)as f32; self.count += 1; } #[cfg(collect_measures)] pub fn merge(&mut self, m: &mut Class) { self.mean = ( self.mean * self.count as f32 + ((m.mean) * m.count as f32 )) / ( self.count + m.count) as f32; self.count += m.count; self.measures.append(&mut m.measures); } #[cfg(not(collect_measures))] pub fn merge(&mut self, m: &mut Class) { self.mean = ( self.mean * self.count as f32 + ((m.mean) * m.count as f32 )) / ( self.count + m.count) as f32; self.count += m.count; } } #[test] fn test_Measure() { let a = Measure::new(1f32,3f32); let b = Measure::new(4f32,5f32); let add = a + b; assert!( add.x == 5f32 && add.y == 8f32 ); let dif = a - b; assert!( dif.x == -3f32 && dif.y == -2f32 ); let mlt = &a * 3f32; assert!( mlt.x == 3f32 && mlt.y == 9f32 ); let div = &b / 2f32; assert!( div.x == 2f32 && div.y == 2.5f32 ); } #[test] fn test_Class() { let mut cls = Class::new( Measure::new(1f32,2f32) ); assert!( cls.mean.x == 1f32 && cls.mean.y == 2f32 ); assert!( cls.count == 1 ); cls.append( Measure::new(3f32,4f32) ); assert!( cls.mean.x == 2f32 && cls.mean.y == 3f32 ); assert!( cls.count == 2 ); } struct Classifier { list:Vec<RefCell<Class>>, ncls_dist:f32 } impl Classifier{ pub fn new(mdist:f32) -> Classifier{ Classifier{ list:Vec::new(), ncls_dist:mdist } } pub fn classificate(&mut self, m: Measure){ let mut min_dist:f32 = f32::MAX; let mut near_cls = 0; let mut index:usize = 0; for i in self.list.iter() { let d = dist( &m, & i.borrow().mean); if d < min_dist { min_dist = d; near_cls = index; } index+=1; } if min_dist < self.ncls_dist{ self.list[near_cls].borrow_mut().append(m); } else { self.list.push(RefCell::new( Class::new(m))); } } #[allow(dead_code)] pub fn merge_classes(&mut self) { let mut uniq: Vec<RefCell<Class>>= Vec::new(); for cls in self.list.iter(){ let mut is_uniq = true; for trg in uniq.iter(){ if dist( &cls.borrow().mean, &trg.borrow().mean) < self.ncls_dist { trg.borrow_mut().merge(&mut *cls.borrow_mut()); is_uniq = false; break; } } if is_uniq { uniq.push(RefCell::new(cls.borrow_mut().clone())); } } self.list = uniq; } } fn read_measures_from_file( name: String ) -> Vec<Measure> { let mut list:Vec<Measure> = Vec::new(); let f = File::open(name).unwrap(); let mut reader = BufReader::new(&f); let mut ret_string = String::new(); while let Ok(size ) = reader.read_line( &mut ret_string){ if size > 0 { let len = ret_string.len() - 1; ret_string.truncate(len); let buffer:Vec<&str> = ret_string.split(' ').collect(); let var:f32 = (buffer[0]).parse::<f32>().unwrap(); let var2:f32 = (buffer[1]).parse::<f32>().unwrap(); list.push(Measure::new(var,var2)); } else{ break; } ret_string.clear(); } return list; } fn main() { let measures = read_measures_from_file(env::args().skip(1).next().unwrap()); let mut clsf = Classifier::new(3f32); let d = Duration::span(||{ for i in measures{ clsf. classificate(i); } if cfg!(merge_classes){ clsf.merge_classes(); } }); println!("work time: {}", d.secs() as f64 + d.extra_nanos()as f64 / 1000000000.0f64); for i in clsf.list{ let i = i.borrow(); println!("[{}, {}]: {}",i.mean.x,i.mean.y, i.count); } }
      
      

        
        
        
      

    
        
        
        
      
      

        
        
        
      

    
     

 #include <vector> #include <cmath> #include <cfloat> #include <iostream> #include <fstream> #include <sstream> #include <string> #include <cassert> #include <ctime> using namespace std; //#define COLLECT_MEASURES //#define MERGE_CLASSES class Measure { public: float x, y; Measure(): x(0), y(0) {} Measure( float X, float Y ): x(X), y(Y) {} Measure( const Measure& m ): x(mx), y(my) {} Measure operator+( const Measure& m ) const { return Measure( x + mx, y + my ); } Measure operator-( const Measure& m ) const { return Measure( x - mx, y - my ); } Measure operator*( float v ) const { return Measure( x * v, y * v ); } Measure operator/( float v ) const { return Measure( x / v, y / v ); } }; void test_Measure() { Measure a(1,3); Measure b(4,5); auto add = a + b; assert( add.x == 5 && add.y == 8 ); auto dif = a - b; assert( dif.x == -3 && dif.y == -2 ); auto mlt = a * 3; assert( mlt.x == 3 && mlt.y == 9 ); auto div = b / 2; assert( div.x == 2 && div.y == 2.5 ); } inline float sqr( float v ) { return v * v; } float dist( const Measure& a, const Measure& b ) { return sqrt( sqr(ax - bx) + sqr(ay - by) ); } class Class { public: Measure mean; size_t N; #ifdef COLLECT_MEASURES vector<Measure> measures; #endif Class( const Measure& m ): mean(m) { N = 1; #ifdef COLLECT_MEASURES measures.push_back(m); #endif } void append( const Measure& m ) { mean = ( mean * N + m ) / ( N + 1 ); N++; #ifdef COLLECT_MEASURES measures.push_back(m); #endif } void merge( const Class& m ) { mean = ( mean * N + m.mean * mN ) / ( N + mN ); N += mN; #ifdef COLLECT_MEASURES measures.insert(measures.end(), m.measures.begin(), m.measures.end()); #endif } }; void test_Class() { auto cls = Class( Measure(1,2) ); assert( cls.mean.x == 1 && cls.mean.y == 2 ); assert( cls.N == 1 ); cls.append( Measure(3,4) ); assert( cls.mean.x == 2 && cls.mean.y == 3 ); assert( cls.N == 2 ); } class Classifier { public: vector<Class*> list; float ncls_dist; Classifier( float mdist ): ncls_dist(mdist) {} void classificate( const Measure& m ) { float min_dist = FLT_MAX; Class* near_cls; for( auto i = list.begin(); i != list.end(); ++i ) { float d = dist( m, (*i)->mean ); if( d < min_dist ) { min_dist = d; near_cls = *i; } } if( min_dist < ncls_dist ) near_cls->append(m); else list.push_back( new Class(m) ); } void mergeClasses() { vector<Class*> uniq; l: for( auto cls = list.begin(); cls != list.end(); ++cls ) { bool is_uniq = true; for( auto trg = uniq.begin(); trg != uniq.end(); ++trg ) { if( dist( (*cls)->mean, (*trg)->mean ) < ncls_dist ) { (*trg)->merge( **cls ); delete (*cls); is_uniq = false; } if( !is_uniq ) break; } if( is_uniq ) uniq.push_back( *cls ); } list = uniq; } ~Classifier() { for( auto i = list.begin(); i != list.end(); ++i ) delete *i; } }; vector<Measure> readMeasuresFromFile( char* name ) { ifstream file( name ); vector<Measure> list; for( string line; getline(file, line); ) { istringstream in( line ); Measure tmp; in >> tmp.x >> tmp.y; list.push_back( tmp ); } return list; } void runTests() { test_Measure(); test_Class(); } int main( int argc, char* args[] ) { //runTests(); auto measures = readMeasuresFromFile( args[1] ); clock_t start = clock(); auto clsf = new Classifier(3); for( auto i = measures.begin(); i != measures.end(); ++i ) clsf->classificate( *i ); #ifdef MERGE_CLASSES clsf->mergeClasses(); #endif clock_t end = clock(); cout << "work time: " << double(end - start) / CLOCKS_PER_SEC << endl; for( auto i = clsf->list.begin(); i != clsf->list.end(); ++i ) cout << "[" << (*i)->mean.x << ", " << (*i)->mean.y << "]: " << (*i)->N << endl; delete clsf; }
      
      

        
        
        
      

    
        
        
        
      
      

        
        
        
      

    
     

 import std.stdio; import std.string; import std.exception; import std.random; import std.math; double normal( double mu=0.0, double sigma=1.0 ) { static bool deviate = false; static float stored; if( !deviate ) { double max = cast(double)(ulong.max - 1); double dist = sqrt( -2.0 * log( uniform( 0, max ) / max ) ); double angle = 2.0 * PI * ( uniform( 0, max ) / max ); stored = dist * cos( angle ); deviate = true; return dist * sin( angle ) * sigma + mu; } else { deviate = false; return stored * sigma + mu; } } struct vec { float x, y; } vec randomVec( in vec m, in vec s ) { return vec( normal(mx, sx), normal(my, sy) ); } auto generate( size_t[vec] classes ) { vec[] ret; foreach( pnt, count; classes ) { auto tmp = new vec[]( count ); foreach( i, ref t; tmp ) t = randomVec( pnt, vec(1,1) ); ret ~= tmp; } return ret; } import std.getopt; void main( string[] args ) { uint k; getopt( args, "count-multiplier|c", &k ); enforce( args.length == 2, format( "wrong number of arguments: use %s <output_file>", args[0] ) ); auto f = File( args[1], "w" ); scope(exit) f.close(); auto vs = generate( [ vec(-8,8): 20 * k, vec(4,0) : 10 * k, vec(-6,-8) : 15 * k ] ); foreach( v; vs.randomSample(vs.length) ) f.writeln( vx, " ", vy ); }
      
      

        
        
        
      

    
        
        
        
      
      

        
        
        
      

    
     

 g++ -O2 -std=c++11 cls.cpp -o cls_cpp && echo "c++ g++ builded" dmd -O -release cls.d -ofcls_d_dmd && echo "d dmd builded" ldc2 -O2 -release cls.d -ofcls_d_ldc2 && echo "d ldc2 builded" gdc -O2 cls.d -o cls_d_gdc && echo "d gdc builded" rustc -O cls.rs -o cls_rs && echo "rust rustc builded"
      
      

        
        
        
      

    
        
        
        
      
      

        
        
        
      

    
     

 g++ -O2 -D COLLECT_MEASURES -D MERGE_CLASSES -std=c++11 cls.cpp -o cls_cpp && echo "c++ g++ builded" dmd -O -version=COLLECT_MEASURES -version=MERGE_CLASSES -release cls.d -ofcls_d_dmd && echo "d dmd builded" ldc2 -O2 -d-version COLLECT_MEASURES -d-version MERGE_CLASSES -release cls.d -ofcls_d_ldc2 && echo "d ldc2 builded" gdc -O2 -fversion=COLLECT_MEASURES -fversion=MERGE_CLASSES cls.d -o cls_d_gdc && echo "d gdc builded" rustc -O --cfg collect_measures --cfg merge_classes cls.rs -o cls_rs && echo "rust rustc builded"