k-means.cpp

#include<iostream> 
#include<string_view>
#include<fstream>
#include<utility>
#include<stdexcept>
#include<vector>
#include<cstdio> 
#include<cmath>
#include<tuple>
#include<random>
#include<cassert>
#include<limits>
#include<srook/algorithm/for_each.hpp> // https://github.com/falgon/SrookCppLibraries/blob/master/srook/algorithm/for_each.hpp

struct ploter{
	constexpr ploter(const char* file):file_(file){}
	
	void output(const char* plt_command)noexcept(false)
	{
		using namespace std::string_literals;

		std::FILE* fp=popen("gnuplot","w");
		if(fp==nullptr)throw std::runtime_error("could not start gnuplot properly");
		std::string file_type,plt_cmd=plt_command;
		plt_cmd+="\n";

		std::copy(std::next(std::begin(file_),file_.find(".")+1),std::end(file_),std::back_inserter(file_type));
		std::string first_cmd="set terminal "s+file_type+"\n"s;
		std::string second_cmd="set output \""s+std::string(file_.data())+"\"\n"s;

		std::fputs(first_cmd.c_str(),fp);
		std::fputs(second_cmd.c_str(),fp);
		std::fputs(plt_cmd.c_str(),fp);

		std::fflush(fp);
		pclose(fp);
	}
private:
	const std::string_view file_;
};

using Point=std::pair<double,double>;

struct k_means{
	k_means(const char* file,std::size_t clustering_size):master_vec(std::move(clustering_size)),initialized(false)
	{
		std::ifstream ifs(file);
		if(!ifs.fail()){
			std::string one_point,first,second;
			while(std::getline(ifs,one_point)){
				std::string::const_iterator iter=std::next(std::begin(one_point),one_point.find(","));
				std::copy(std::cbegin(one_point),iter,std::back_inserter(first));
				std::copy(std::next(iter,1),std::cend(one_point),std::back_inserter(second));

				input_data.emplace_back(Point{std::stod(first),std::stod(second)});
				first.clear();
				second.clear();
			}
		}
	}

	template<
		class... Ts,
		std::enable_if_t<(sizeof...(Ts)>0) and std::conjunction_v<std::is_same<Point,Ts>...>,std::nullptr_t> =nullptr
	>
	void set_initial_point(Ts&&... ts)noexcept
	{
		const std::tuple<std::decay_t<Ts>...> tpl{std::forward<Ts>(ts)...};
		assert(sizeof...(ts)==clustering_size());
		srook::for_each(srook::make_counter(tpl),[this](const auto& x,std::size_t i){master_vec[i]=x;});
		initialized=true;
	}

	template<class T>
	using rand_type=
		std::conditional_t<
			!std::is_same_v<std::decay_t<T>,double> and !std::is_same_v<std::decay_t<T>,float>,
			std::uniform_int_distribution<std::decay_t<T>>,
			std::uniform_real_distribution<std::decay_t<T>>
		>;
	template<class X,class Y>
	void set_initial_point(X&& x_min,X&& x_max,Y&& y_min,Y&& y_max)
	{
		std::random_device rd;
		std::mt19937 gen(rd());

		rand_type<X> x_dis(std::move(x_min),std::move(x_max));
		rand_type<Y> y_dis(std::move(y_min),std::move(y_max));
		
		for(auto& v:master_vec)
			v=Point{x_dis(gen),y_dis(gen)};
		initialized=true;
	}

	void clustering()
	{
		if(!initialized){
			std::vector<Point::first_type> xs(input_data.size());
			std::vector<Point::second_type> ys(input_data.size());
			srook::for_each(srook::make_counter(xs),[this](auto& x,std::size_t i){x=input_data[i].first;});
			srook::for_each(srook::make_counter(ys),[this](auto& y,std::size_t i){y=input_data[i].second;});
			
			set_initial_point(
					*std::min_element(std::begin(xs),std::end(xs)),*std::max_element(std::begin(xs),std::end(xs)),
					*std::min_element(std::begin(ys),std::end(ys)),*std::max_element(std::begin(ys),std::end(ys))
			);
		}

		std::vector<int> prev_cluster(input_data.size(),0);
		cluster.resize(input_data.size(),-1);

		while(!std::equal(std::begin(prev_cluster),std::end(prev_cluster),std::begin(cluster),std::end(cluster))){
			prev_cluster=cluster;

			srook::for_each(
					srook::make_counter(input_data),
					[this](const auto& v,std::size_t i)
					{
						int max_cluster=-1;
						double arg_max=std::numeric_limits<double>::infinity();

						srook::for_each(
								srook::make_counter(master_vec),
								[this,&arg_max,&v,&max_cluster](const auto& mp,std::size_t j)
								{
									if(arg_max > distance(v,mp)){
										arg_max=distance(v,mp);
										max_cluster=j;
									}
								}
						);
						cluster[i]=max_cluster;
					}
			);
			
			srook::for_each(
					srook::make_counter(master_vec),
					[this](auto& m,int i)
					{
						int c=0;
						m.first=0;
						m.second=0;
						srook::for_each(
								srook::make_counter(input_data),
								[i,&c,&m,this](const auto& v,std::size_t j)
								{
									if(cluster[j]==i){
										m.first+=v.first;
										m.second+=v.second;
										++c;
									}
								}
						);
						if(c){
							m.first/=c;
							m.second/=c;
						}
					}
			);
		}
	}

	std::vector<Point>::size_type clustering_size()const noexcept
	{
		return master_vec.size();
	}
private:
	std::vector<Point> input_data;
	std::vector<int> cluster;
	std::vector<Point> master_vec;
	bool initialized;
private:
	double distance(const Point& l,const Point& r)noexcept
	{
		return std::sqrt(std::pow(l.first-r.first,2.0) + std::pow(l.second-r.second,2.0));
	}

	friend std::ofstream& operator<<(std::ofstream& ofs,const k_means& km)
	{
		for(std::size_t i=0; i<km.clustering_size(); ++i){
			srook::for_each(
					srook::make_counter(km.input_data),
					[i,&ofs,&km](const auto& v,std::size_t j)
					{
						if(std::size_t(km.cluster[j])==i){
							ofs<<v.first<<" "<<v.second<<"\n";
						}
					}
			);
			ofs<<"\n\n";
		}
		return ofs;
	}

	friend std::ostream& operator<<(std::ostream& os,const k_means& km)
	{
		for(std::size_t i=0; i<km.clustering_size(); ++i){
			srook::for_each(
					srook::make_counter(km.input_data),
					[i,&os,&km](const auto& v,std::size_t j)
					{
						if(std::size_t(km.cluster[j])==i){
							os<<v.first<<","<<v.second<<"\n";
						}
					}
			);
			os<<"\n";
		}
		return os;
	}
};

int main()
{ 
	k_means km("data",3); // 3つにclustering.
//	km.set_initial_point(Point{2.0,90.0},Point{5.0,50.0}); // initial centerを任意の座標に設定.
	km.clustering();
	
	const char* result_file="result";
	std::ofstream ofs(result_file);
	ofs<<km;

	using namespace std::string_literals;

	ploter pl("output.png");
	std::string cmd="plot \""s+std::string(result_file)+"\" index 0,\"\" index 1,\"\" index 2";

	try{
		pl.output(cmd.c_str());
	}catch(const std::runtime_error& exp){
		std::cerr<<exp.what()<<std::endl;
	}
}

LICENSE

MIT License

Copyright (c) 2017 roki

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

something_mouse.py

#!/usr/local/bin/python3
# coding: UTF-8

import random
import math

def circle(f,pl,delimiter,ir,cx,cy):
    for _ in range(pl):
        r = random.random()*math.pi*2.0
        s = ir * math.sqrt(random.random())
        x = s * math.cos(r) + 200 + cx
        y = s * math.sin(r) + 200 + cy
        f.write(str(x)+delimiter+str(y)+'\n')

c1_r = 150
c2_r = 100
c1_plot = 10000
c2_plot = 5000
delimiter = ','

f = open('mouse_data','w')
circle(f,c2_plot,delimiter,c2_r,0,0)
circle(f,c2_plot,delimiter,c2_r,c2_r*2,0)
circle(f,c1_plot,delimiter,c1_r,c2_r,-c2_r*2)

f.close()