深度置信网络

深度置信网络具有深层的网络结构，深度学习工具箱中可以调用DBN的程序。

function test_example_DBN

load mnist_uint8;  % 载入数据

train_x = double(train_x) / 255;   % train_x为784个像素值， 这里是在做数据归一化，（0-1）

test_x  = double(test_x) / 255;

train_y = double(train_y);

test_y  = double(test_y);

 

%%  ex1 train a 100 hidden unit RBM and visualize its weights

% rng(0);

dbn.sizes = [  100  ];  %设置网络隐藏单元数为100

opts.numepochs =   1;       %设置训练迭代次数。

opts.batchsize = 100;     %批次大小

opts.momentum  =   0;    %动量

opts.alpha     =   1;     % 学习率

dbn = dbnsetup(dbn, train_x, opts);  %初始化RBM的参数

dbn = dbntrain(dbn, train_x, opts);   %开始训练

figure; visualize(dbn.rbm{1}.W');   %   RBM 权重

 

%%  ex2 train a 100-100 hidden unit DBN and use its weights to initialize a NN

% rng(0);

%train dbn

dbn.sizes = [100 100];

opts.numepochs =   10;

opts.batchsize = 100;

opts.momentum  =   0;

opts.alpha     =   1;

dbn = dbnsetup(dbn, train_x, opts);

dbn = dbntrain(dbn, train_x, opts);

 

%unfold dbn to nn

nn = dbnunfoldtonn(dbn, 10);           %设计一个有十个输出单元的NN，并用已经训练好的DBN的权值参数去初始化相应结构的NN网络

nn.activation_function = 'sigm';        %设置激活函数为 sigm

 

%train nn

opts.numepochs =  1;                   % 设置训练参数

opts.batchsize = 10;                          

nn = nntrain(nn, train_x, train_y, opts);   %训练网络

[er, bad] = nntest(nn, test_x, test_y);     %测试网络错误率

 

assert(er < 0.10, 'Too big error');