本程序训练了多个神经网络（它们的隐藏层神经元数量不同），并允许用户选择性能最好的网络。

The program trains several neural networks (that vary in their number of hidden layer neurons) and enables users to pick the best. 

该程序分别使用名为Inputs1.txt和Targets1.txt文件中包含的输入和目标数据来训练网络。

The program trains the networks using input and target data contained in files named Inputs1.txt and Targets1.txt respectively. 

程序将提供的数据随机分成3部分：70%用于训练，15%用于验证，15%用于测试。

The program randomly splits the supplied data into 3 portions: 70% for training, 15% for validation, and 15% for testing. 

用户可以选择在第68行到第70行中更改以上数据的比例。

The user has the option to change this proportioning in lines 68 to 70.

用户还可以选择使用自己的测试数据：这些数据应与Inputs2.txt和Targets2.txt相同的格式保存。

The user also has the option to use his/her own test data: these should be saved in same format as Inputs2.txt and Targets2.txt

当程序运行时，将训练100个不同的网络：各网络的隐藏层神经元数量不同。

When the program runs, it trains 100 different networks: the networks differ in their number of hidden layer neurons. 

生成的神经网络保存在名为networks的文件夹中，该文件夹与运行此程序的文件夹相同。

The networks generated are saved in in a folder named networks, within the same folder where this program is run. 

保存为net1的网络有1个隐藏层神经元，net2有2个，net3有3个，net4有4个，…

The network saved as net1 has 1 hidden layer neuron, net2 has 2, net3 has 3, net4 has 4,....

用户可以将第65行中的值1:100更改为所需的隐藏层神经元个数。

Users may change the values 1:100 in line 65 to their desired values.

trainlm采用了Levenberg-Marquardt算法进行训练。

Training is done using the Levenberg-Marquardt algorithm, trainlm. 

用户也可以通过修改第62行来更改自己需要的算法。

Users may change to their desired algorithm by changing line 62.

该程序还将计算得到均方根误差（RMSEs）保存在一个名为rmse.txt的文件中。

The program also computes and saves root-mean-squared-errors (RMSEs) in a file named rmse.txt. 

文件中的数据分为3列：第1列为隐藏层神经元的数量，第2列为根据随机15%测试数据计算的RMSE，第3列为根据用户自己的测试数据计算的RMSE。

Data in the file is in 3 columns: column 1 for the number of hidden layer neurons, column 2 for the RMSE computed on the random 15% test data, and column 3 for the RMSE computed on the user's own test data.

此文件上的数据将指导用户决定哪种神经网络的性能最好。

The data on this file guides the user to decide which of the networks is best. 

具有较小RMSE的网络性能更好，尤其是对于根据用户自己的测试数据计算的RMSE，这些数据超出了用于训练的数据覆盖范围。

Networks with smaller RMSEs are better, especially for the RMSEs computed on the user's own test data which is outside the range of data used for the training. 

关于如何使用RMSE来选择最优网络的更详细的指南包含在本程序作者编写的一本书中，书名为“Computer Neural Networks on MATLAB”。

More detailed guide on how to use the RMSEs to choose an optimal network is contained in a book authored by the writer of this program and titled "Computer Neural Networks on MATLAB"。

完整源码下载地址：

http://page2.dfpan.com/fs/5lcej2c21929116e3a7/  

更多精彩文章请关注微信号：