基于softmax回归的MNIST数据集手写数字识别matlab仿真+仿真录像

clc; clear; close all; warning off; addpath(genpath(pwd)); %--加载数据----------------------------------------% load 'PreTreatedData/trainImages.mat'; load 'PreTreatedData/trainLabels.mat'; load 'PreTreatedData/testImages.mat'; load 'PreTreatedData/testLabels.mat'; %--训练---------------------------------------------------% %iter = 0;% 迭代次数 trainPara.lossMax = 0.01; % 最大损失率 trainPara.iterMax = 2000; % 最大训练次数 %trainPara.alpha = ones(c+1,1).*0.001; % 迭代步长 trainPara.alpha = 0.1; % 学习率 trainPara.lossLimt=0.01; %等步长到变步长的切换界限 trainPara.eta = 0.618; %迭代学习率变化率 trainPara.num2Train=1000; %选取要训练的图片数量 %trainImage.data每一行代表一张图片 [r,c]=size(trainImage.data(1:trainPara.num2Train,:)); %归一化，选取10000张图片，X的拓展矩阵,同时转置 X_hat=[trainImage.data(1:trainPara.num2Train,:)./255 ones(r,1)]'; y=trainLabel.data(1:trainPara.num2Train)+1; %将值偏移到1-10 numClasses=10; theta0=[ones(c,numClasses)./c;zeros(1,numClasses)]; %模型权重初值 softmaxModel=funcSoftmaxTrain(@funcdCost,@funcCost,trainPara,theta0,numClasses,X_hat,y); figure; xplot=(1:trainPara.iterMax); plot(xplot,softmaxModel.loss); %set(gcf,'unit','centimeters','position',[20 5 7 6]); xlabel('numOfIter'); ylabel('Loss'); xlim([0 trainPara.iterMax]); ylim([0 1.6]); %--测试---------------------------------------------------% [r_test,c_test]=size(testImage.data); y_test=testLabel.data+1; %训练集列表，偏移1 X_hat_test=[testImage.data./255 ones(r_test,1)]'; %归一化，获取拓展矩阵,转置 accuracy=funcSoftmaxTest(softmaxModel.theta,numClasses,X_hat_test,y_test); %进入测试 disp(['识别准确率为',num2str(accuracy*100),'%']);