[人工智能]基于MATLAB的BR神经网络交叉验证实践

基于MATLAB的BR神经网络交叉验证实践

在做项目需要对BR神经网络进行K-Flod交叉验证并最终得到结果，于是写了这个代码。但是，有疑问的一点是这行代码：

net1 = feedforwardnet(3, ‘trainbr’);

不确定“net1”应该放在“for”循环的外面还是里面，放在不同的位置会得到不同的结果。

// Author LJS
clear

load allData.mat;

indices = crossvalind('KFold',200,5);
net1 = feedforwardnet(5, 'trainbr');
k=0;
record = zeros(1,5);
record2 = zeros(1,5);
record3 = zeros(1,5);
for i=1:5
    k = k+1;
	test1=(indices==k); %结果为逻辑值，每次循环选取一个组作为测试集
	t_rain1=~test1; %取test的补集即为训练集
	Train_Data=allData(t_rain1,:); %提取训练集数据 
	TestData=allData(test1,:); %提取测试集数据
    
    Feature = Train_Data(:,1:13);
    test_Feature = TestData(:,1:13);
    DPP = Train_Data(:,14);
    test_DPP = TestData(:,14);
    
    Feature = Feature';
    test_Feature = test_Feature';
    DPP = DPP';
    test_DPP = test_DPP';
    
    net1.divideFcn = 'dividerand';
    net1.divideParam.testRatio = 0;
    net1.divideParam.valRatio = 0;

    [nets,tr] = train(net1, Feature, DPP);

    predict = nets(test_Feature);
    perf = mse(nets,test_DPP,predict);
    
    record(i) = perf;
    
end

mean_mse = mean(record);
rmse = sqrt(mean_mse);

交叉验证的部分代码参考了 沙子在流 的博文。

引用链接：
https://blog.csdn.net/weixin_38672958/article/details/116453688?spm=1001.2014.3001.5501