【读书1】【2017】MATLAB与深度学习——SGD方法的实现(1)

通过对训练输出与正确输出的比较，我们可以看出神经网络是如何进行训练的。

We can see how well the neural network wastrained by comparing the output with the correct output.

clear all
X = [ 0 0 1;
0 1 1;
1 0 1;
1 1 1;
];
D = [ 0 0 1 1 ];

W = 2*rand(1, 3) - 1; % 随机数初始化权值
for epoch = 1:10000 % train

W = DeltaSGD(W,X, D);
end
N = 4; % inference
for k = 1:N

x = X(k, ?’;

v = W*x;

y = Sigmoid(v)
end

该代码用-1到1之间的随机实数初始化权值。

This code initializes the weights withrandom real numbers between -1 and 1.

执行此代码会输出以下结果。

Executing this code produces the followingvalues.

这些输出值与D中的正确输出非常接近。

These output values are very close to thecorrect outputs in D.

因此，我们可以得出结论，神经网络已经得到适当的训练。

Therefore, we can conclude that the neuralnetwork has been properly trained.

本书中的每一个示例代码都是在单独的文件中实现算法和测试程序。

Every example code in this book consists ofthe implementation of the algorithm and the test program in separate files.

这是因为将它们放在一起往往会使代码更加复杂，从而阻碍我们对算法的有效分析。

This is because putting them together oftenmakes the code more complicated and hampers efficient analysis of thealgorithm.

测试程序的文件名以Test开头，紧跟着的是算法文件名。

The file name of the test program startswith Test and is followed by the name on the algorithm file.

算法文件是根据函数名称命名的，符合MATLAB的命名约定。

The algorithm file is named after thefunction name, in compliance with the naming convention of MATLAB.

例如，DeltaSGD函数的实现文件名称为DeltaSGD.m。

For example, the implementation file of theDeltaSGD function is named DeltaSGD.m.

算法实现命名举例：DeltaSGD.m

Algorithm implementation example/DeltaSGD.m

测试程序命名举例：TestDeltaSGD.m

Test program example/ TestDeltaSGD.m

批处理方法的实现（Implementation of the Batch Method）

函数DeltaBatch使用批处理方法实现方程2.7的增量规则。

The function DeltaBatch implements thedelta rule of Equation 2.7 using the batch method.

它采用神经网络的权值和训练数据，并返回训练后的权值。

It takes the weights and training data ofthe neural network and returns trained weights.

W = DeltaBatch(W, X, D)

在这个函数的定义中，所有变量与函数DeltaSGD中的变量具有相同的含义；W是神经网络的权重，X和D分别是训练数据的输入和正确输出。

In this function definition, the variablescarry the same meaning as those in the function DeltaSGD; W is the weight ofthe neural network, X and D are the input and correct output of the trainingdata, respectively.

下面给出了实现函数DeltaBatch的DeltaBatch.m文件。

The following listing shows theDeltaBatch.m file, which implements the function DeltaBatch.

function W = DeltaBatch(W, X, D)

alpha = 0.9;

dWsum = zeros(3,1);

N = 4;

for k = 1:N

x = X(k, ?’;
d = D(k);
v = Wx;
y = Sigmoid(v);
e = d - y;
delta = y(1-y)e;
dW = alphadelta*x;
dWsum = dWsum + dW;

end
dWavg = dWsum / N;

W(1) = W(1) + dWavg(1);
W(2) = W(2) + dWavg(2);
W(3) = W(3) + dWavg(3);
end

——本文译自Phil Kim所著的《Matlab Deep Learning》

更多精彩文章请关注微信号：