线性回归梯度下降matlab实现

最新推荐文章于 2023-01-31 15:43:39 发布

Lee_zix

最新推荐文章于 2023-01-31 15:43:39 发布

阅读量998

点赞数

分类专栏：机器学习

本文链接：https://blog.csdn.net/Lee_zix/article/details/44889979

版权

机器学习专栏收录该内容

3 篇文章 0 订阅

订阅专栏

接中午的理论部分

clc;
clear;
%x0 = 1,h(x) = x0*sita0 + x1*sita1+........
x = [1 1.15
    1,1.9
    1,3.06
    1,4.66
    1,6.84
    1,7.95];
y = [1.37,2.4,3.02,3.06,4.22,5.42];
subplot(1,2,1);
plot(x(:,2),y);
hold on;
%×î´óµü´ú´ÎÊý
loop_max = 10000;
%ÊÕÁ²µÄ¾«¶È
epsilon = 0.0001;
%µü´ú²½³¤
alpha =0.001;
%
diff  = 0;
lastsita0 = 0;  
lastsita1 = 0;
m = size(x,1);
sita = [0,0];
count = 0;
finish = 0;
n = input('choose the method of gradient descent:1.incremental gradient descent  2.batch gradient descent\n');
switch(n)
    case(1)
%ÔöÁ¿ÌÝ¶ÈÏÂ½µ
        while count<loop_max
             count = count+1
            for i = 1:1:m
            diff = sita(1)*x(i,1) + sita(2)*x(i,2) - y(i);
            sita(1) = sita(1) - alpha * diff*x(i,1);
            sita(2) = sita(2) - alpha * diff*x(i,2);
            end
            %ÅÐ¶ÏÊÇ·ñÐèÒª¼ÌÐøµü´ú
             if(abs(sita(1) - lastsita0)<epsilon && abs(sita(2) - lastsita1)<epsilon)
               finish = 1;
             else
                lastsita0 = sita(1);
                lastsita1 = sita(2);
                sita
                diff
                subplot(1,2,2);
                line(count,diff,'Marker','.');
                ylabel('×îÐ¡¾ù·½');
                xlabel('µü´ú´ÎÊý');
                hold on;
             end

            if (finish ==1)
              break;
            end
        end%end of while

%ÅúÌÝ¶ÈÏÂ½µ
        case (2)
            sum = 0;
            while count<loop_max
                count = count+1
                for i = 1:1:m
                    diff = sita(1)*x(i,1) + sita(2)*x(i,2) - y(i);
                    sum= sum +diff*x(i,1);
                end
                sita(1) = sita(1)- alpha*sum;
                sum =0;
                for i = 1:1:m
                     diff = sita(1)*x(i,1) + sita(2)*x(i,2) - y(i);
                     sum= sum +diff*x(i,2);
                end
                sita(2) = sita(2)- alpha*sum;
                if(abs(sita(1) - lastsita0)<epsilon && abs(sita(2) - lastsita1)<epsilon)
                    finish = 1;
                else
                     lastsita0 = sita(1);
                     lastsita1 = sita(2);
                     sita
                     diff
                     subplot(1,2,2);
                     line(count,diff,'Marker','.');
                     ylabel('×îÐ¡¾ù·½');
                     xlabel('µü´ú´ÎÊý');
                     hold on;
                end

            if (finish ==1)
              break;
            end
           end%end of while
end%end of switch


subplot(1,2,1);
plot(x(:,2),sita(1)+sita(2)*x(:,2),'r')

注释全乱码了。。。。–！
准备用线性回归来拟合y = 1+0.5x。
分别用了增量梯度下降和批梯度下降来计算LMS的最小值。理想的真实值当然是这里写图片描述 1=1，2=0.5，增量梯度下降方法和批梯度下降方法个选用了步长为0.001,0.005，0.0001三种情况，发现学习速率对参数确定的结果的影响很大，学习速率太大太小都不能得到准确的结果。下面是三种情况的结果显示图