模式识别技术-非线性函数拟合问题-数据集训练（问题五）

sym x%建立符号x %引入新的抽样点数 kl = [];%用于检测n值的中间变量 for p = 1:6 n = [15 20 25 50 100 200]; J = n(p);%设定新的抽样点数 kl(end+1) = J; MSE_1_10 = zeros(10,100); for i = 1:10 if i == 1 for j = 1:100 %建立训练样本和测试样本 Random_row = randperm(1000,J); TrainData = hw1p5data(Random_row,:); TestData = hw1p5data(); TestData(Random_row,:) = []; %利用最小二乘法进行拟合训练 x_train = TrainData(:,1); y_train = TrainData(:,2); x_text = TestData(:,1); y_truth = TestData(:,2);%y的真实值 Parameter = polyfit(x_train,y_train,1); y_test = Parameter(1) .* x_text + Parameter(2);%y的测试值 %计算均方误差 N = 1000-J;%求解数据量 Error = y_truth - y_test; MSE = dot(Error,Error)/N; MSE_1_10(sub2ind(size(MSE_1_10), i, j)) = MSE; end else for k = 1:100 %建立训练样本和测试样本 Random_row = randperm(1000,J); TrainData = hw1p5data(Random_row,:); TestData = hw1p5data(); TestData(Random_row,:) = []; %利用最小二乘法进行拟合训练 x_train = TrainData(:,1); y_train = TrainData(:,2); x_text = TestData(:,1); y_truth = TestData(:,2);%y的真实值 Parameter = polyfit(x_train,y_train,i); %利用polyval函数计算y_test y_test = polyval(Parameter,x_text); %计算均方误差 N = 1000-J;%求解数据量 Error = y_truth - y_test; MSE = dot(Error,Error)/N; MSE_1_10(sub2ind(size(MSE_1_10), i, k)) = MSE; end end end %求100次操作的均值 MSE_M = []; for i = 1:10 MSE_1 = mean(MSE_1_10(i,:)); MSE_M(end+1) = MSE_1; end MSE_log = log(MSE_M); %取对数 %绘制log(MSE)与阶数的关系 n = 1:1:10; xlabel('拟合的阶数'); ylabel('logMSE'); subplot(2,3,p),plot(n,MSE_log),title({'logMSE与拟合阶数的关系', ['拟合阶数 = ', num2str(J)] }); end