# liblgd

**Repository Path**: yizhangss/liblgd

## Basic Information

- **Project Name**: liblgd
- **Description**: C++ library of the Lévy-Gradient Descent (L-GD) method
- **Primary Language**: C++
- **License**: GPL-3.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-09-10
- **Last Updated**: 2025-02-01

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# liblgd

#### 介绍
C++ library of the Lévy-Gradient Descent (L-GD) method

#### 软件架构
软件架构说明


#### 安装教程

算法库默认使用 Cmake 编译工具进行编译，可在不同平台生成相应的可执行或工程文件。请用户自行下载安装 Cmake 软件后按如下方法进行编译。此方法适用于 MacOS 或 Linux 系统 （默认的编译器为 gcc，用户可在 CMakeLists.txt 中自行修改编译器与安装地址），Windows 用户请使用 VS studio 等编译工具新建项目并拷贝 src/lib 文件夹下所有文件至新项目并编译动态库。

```shell
mkdir build
cd build
cmake ..
make
make install
```

#### 使用说明

1.  xxxx
2.  xxxx
3.  xxxx

#### Examples

```cpp
#include "lgd.h"
#include "iostream"
#include "cmath"
#include "vector"
#include "string"

#define pi (atan(1.0)*4.0)

enum gradient_type_e
{
	Dnull,
	Dx,
	Dy
};

struct gaussian_para
{
	double mu_x, mu_y, sigma_x, sigma_y, rho;
};

double gaussian_distribution(double x, double y, const gaussian_para &p, 
	gradient_type_e mode = Dnull)
{
	double part = -0.5*(pow((x-p.mu_x)/p.sigma_x,2) -2*p.rho*(x-p.mu_x)*(y-p.mu_y)/(p.sigma_x*p.sigma_y)
		+ pow((y-p.mu_y)/p.sigma_y,2)) / (1.0-p.rho*p.rho);

	double part2;
	if (mode == Dnull)
	{
		return exp(part)/(2*pi*p.sigma_x*p.sigma_y*sqrt(1-p.rho*p.rho));
	}
	else if (mode == Dx)
	{
		part2 = -1.0*((x-p.mu_x)/(p.sigma_x*p.sigma_x) - p.rho*(y-p.mu_y)/(p.sigma_x*p.sigma_y))/(1.0-p.rho*p.rho);
		return part2*exp(part)/(2*pi*p.sigma_x*p.sigma_y*sqrt(1-p.rho*p.rho));
	}
	else if (mode == Dy)
	{
		part2 = -1.0*((y-p.mu_y)/(p.sigma_y*p.sigma_y) - p.rho*(x-p.mu_x)/(p.sigma_x*p.sigma_y))/(1.0-p.rho*p.rho);
		return part2*exp(part)/(2*pi*p.sigma_x*p.sigma_y*sqrt(1-p.rho*p.rho));
	}
	else
	{
		std::string err_str = "Invalid calculation mode.";
		throw err_str;
	}
}

static std::vector<gaussian_para> para;

double evaluate(void *instance, const double *x, double *g, const int n)
{
	double fx = 0.0;
	g[0] = g[1] = 0.0;

	for (int i = 0; i < para.size(); i++)
	{
		fx += -1.0*gaussian_distribution(x[0], x[1], para[i]);
		g[0] += -1.0*gaussian_distribution(x[0], x[1], para[i], Dx);
		g[1] += -1.0*gaussian_distribution(x[0], x[1], para[i], Dy);
	}

	fx += 12.82906044;

	g[0] *= 2.0*fx;
	g[1] *= 2.0*fx;
	return fx*fx;
}

double evaluate_fx_only(void *instance, const double *x, const int n)
{
	double fx = 0.0;

	for (int i = 0; i < para.size(); i++)
	{
		fx += -1.0*gaussian_distribution(x[0], x[1], para[i]);
	}

	fx += 12.82906044;

	return fx*fx;
}

int progress(void *instance, lgd_float best_fx, const lgd_float *best_x, 
	const lgd_float *best_g, const lgd_float *x, const lgd_float *g, const lgd_para *param, 
	const int n_size, const int curr_k, const int best_k)
{
	std::cerr << "fight time: " << curr_k << ", best fx: " << best_fx << ", found at fight time: " << best_k << std::endl;
	std::cerr << "\033[1A\033[K";
	return 0;
}

int main(int argc, char *argv[])
{
	gaussian_para tmp_p;

	tmp_p.mu_x = 0.50; tmp_p.mu_y = 0.40; tmp_p.sigma_x = 0.15; tmp_p.sigma_y = 0.10; tmp_p.rho = 0.2;
	para.push_back(tmp_p);

	tmp_p.mu_x = 0.60; tmp_p.mu_y = 0.70; tmp_p.sigma_x = 0.10; tmp_p.sigma_y = 0.20; tmp_p.rho = 0.0;
	para.push_back(tmp_p);

	lgd_float fx;
	lgd_float x[2] = {0.25, 0.20};
	lgd_float low[2] = {0.0, 0.0};
	lgd_float hig[2] = {1.0, 1.0};

	lgd_para my_para = lgd_default_parameters();
	my_para.fight_times = 2000;
	my_para.alpha = 0.01;
	my_para.mu = 0.001;

	lgd_float x_mean[2] = {0.0, 0.0};
	lgd_float x_stddev[2] = {0.0, 0.0};

	int ret = lgd_solver_dynamic(&x_mean[0], &x_stddev[0], evaluate, progress, &fx, &x[0], 2, &my_para, nullptr, 
		&low[0], &hig[0]);
	//int ret = lgd_solver_pseudo_dynamic(&x_mean[0], &x_stddev[0], evaluate_fx_only, progress, &fx, &x[0], 2, &my_para, nullptr, 
	//	&low[0], &hig[0]);
	std::cerr << lgd_error_str(ret) << std::endl;

	std::clog << "best-fx = " << fx << std::endl;
	std::clog << "best-model: " << x[0] << " " << x[1] << std::endl;
	std::clog << "mean-model: " << x_mean[0] << " " << x_mean[1] << std::endl;
	std::clog << "mean-stddev: " << x_stddev[0] << " " << x_stddev[1] << std::endl;
	std::clog << "x-range: " << x_mean[0] - x_stddev[0] << " " << x_mean[0] + x_stddev[0] << std::endl;
	std::clog << "y-range: " << x_mean[1] - x_stddev[1] << " " << x_mean[1] + x_stddev[1] << std::endl;
	return 0;
}
```

#### 参与贡献

1.  Fork 本仓库
2.  新建 Feat_xxx 分支
3.  提交代码
4.  新建 Pull Request


#### 码云特技

1.  使用 Readme\_XXX.md 来支持不同的语言，例如 Readme\_en.md, Readme\_zh.md
2.  码云官方博客 [blog.gitee.com](https://blog.gitee.com)
3.  你可以 [https://gitee.com/explore](https://gitee.com/explore) 这个地址来了解码云上的优秀开源项目
4.  [GVP](https://gitee.com/gvp) 全称是码云最有价值开源项目，是码云综合评定出的优秀开源项目
5.  码云官方提供的使用手册 [https://gitee.com/help](https://gitee.com/help)
6.  码云封面人物是一档用来展示码云会员风采的栏目 [https://gitee.com/gitee-stars/](https://gitee.com/gitee-stars/)