public static void Excute()
{
double[][] inputs;
double[][] outputs;
double[][] testInputs;
double[][] testOutputs;
// Load ascii digits dataset.
inputs = DataManager.Load(@"data.txt", out outputs);
// The first 500 data rows will be for training. The rest will be for testing. 第一个500数据用来训练，剩下的用来测试
testInputs = inputs.Skip(500).ToArray();
testOutputs = outputs.Skip(500).ToArray();
inputs = inputs.Take(500).ToArray();
outputs = outputs.Take(500).ToArray();
// Setup the deep belief network and initialize with random weights.
DeepBeliefNetwork network = new DeepBeliefNetwork(inputs.First().Length, 10, 10);
new GaussianWeights(network, 0.1).Randomize();
network.UpdateVisibleWeights();
// Setup the learning algorithm.
DeepBeliefNetworkLearning teacher = new DeepBeliefNetworkLearning(network)
{
Algorithm = (h, v, i) => new ContrastiveDivergenceLearning(h, v)
{
LearningRate = 0.1,
Momentum = 0.5,
Decay = 0.001,
}
};
// Setup batches of input for learning.
int batchCount = System.Math.Max(1, inputs.Length / 100);
// Create mini-batches to speed learning.
int[] groups = Accord.Statistics.Tools.RandomGroups(inputs.Length, batchCount);
double[][][] batches = inputs.Subgroups(groups);
// Learning data for the specified layer.
double[][][] layerData;
// Unsupervised learning on each hidden layer, except for the output layer.
for (int layerIndex = 0; layerIndex < network.Machines.Count - 1; layerIndex++)
{
teacher.LayerIndex = layerIndex;
layerData = teacher.GetLayerInput(batches);
for (int i = 0; i < 200; i++)
{
double error = teacher.RunEpoch(layerData) / inputs.Length;
if (i % 10 == 0)
{
Console.WriteLine(i + ", Error = " + error);
}
}
}
// Supervised learning on entire network, to provide output classification.
var teacher2 = new Neuro.Learning.BackPropagationLearning(network)
{
LearningRate = 0.1,
Momentum = 0.5
};
// Run supervised learning.
for (int i = 0; i < 500; i++)
{
double error = teacher2.RunEpoch(inputs, outputs) / inputs.Length;
if (i % 10 == 0)
{
Console.WriteLine(i + ", Error = " + error);
}
}
// Test the resulting accuracy.
int correct = 0;
for (int i = 0; i < inputs.Length; i++)
{
double[] outputValues = network.Compute(testInputs[i]);
if (DataManager.FormatOutputResult(outputValues) == DataManager.FormatOutputResult(testOutputs[i]))
{
correct++;
}
}
Console.WriteLine("Correct " + correct + "/" + inputs.Length + ", " + System.Math.Round(((double)correct / (double)inputs.Length * 100), 2) + "%");
Console.Write("Press any key to quit ..");
Console.ReadKey();
}
public static void Excute2()
{
double[][] inputs;
double[][] outputs;
double[][] testInputs;
double[][] testOutputs;
// Load ascii digits dataset.
inputs = DataManager.Load(@"data.txt", out outputs);
// The first 500 data rows will be for training. The rest will be for testing. 第一个500数据用来训练，剩下的用来测试
testInputs = inputs.Skip(500).ToArray();
testOutputs = outputs.Skip(500).ToArray();
inputs = inputs.Take(500).ToArray();
outputs = outputs.Take(500).ToArray();
// Setup the deep belief network and initialize with random weights. 设置深度神经网络和初始化随机砝码
DeepBeliefNetwork network = new DeepBeliefNetwork(inputs.First().Length, 10, 10); //网络的输入数量Length 每个层中隐藏的神经元的数量10, 10
new GaussianWeights(network, 0.1).Randomize(); //高斯砝码 使用标准偏差。一般值在0.001—0.1范围内。 默认值为0.1。
//Randomize 使用高斯分布的网络的权重
network.UpdateVisibleWeights(); //通过复制隐藏层中权重的反向来更新可见层的权重。
// Setup the learning algorithm. 设置学习法则
DeepBeliefNetworkLearning teacher = new DeepBeliefNetworkLearning(network);
//自定义神经网络法则
// 设置用于指定和创建深度网络的每个层的学习算法的配置函数。Algorithm
teacher.Algorithm = (h, v, i) => {
return(new ContrastiveDivergenceLearning(h, v)
{
LearningRate = 0.1, //学习速率
Momentum = 0.5, //动力
Decay = 0.001, //腐烂
});
};
//teacher.Algorithm = (h, v, i) => new ContrastiveDivergenceLearning(h, v)
//{
// LearningRate = 0.1,
// Momentum = 0.5,
// Decay = 0.001,
//};
// Setup batches of input for learning.
int batchCount = System.Math.Max(1, inputs.Length / 100);//设置学习次数
// Create mini-batches to speed learning.
int[] groups = Accord.Statistics.Classes.Random(inputs.Length, batchCount); //创建小批量 速度学习
double[][][] batches = inputs.Separate(groups); //分离
// Learning data for the specified layer.
double[][][] layerData; //为指定层 学习数据
// Unsupervised learning on each hidden layer, except for the output layer.除了输出层之外，在每个隐藏层上进行无监督学习。
//network.Machines.Count 在这个深网络的每一层上得到受限制的玻尔兹曼机器。
for (int layerIndex = 0; layerIndex < network.Machines.Count - 1; layerIndex++)
{
teacher.LayerIndex = layerIndex;
/*
* 获取训练数据所需的学习数据。
* 这个函数的返回应该被传递给No.Posial.SurvivyFieldWorksPr.RunEpoch（System，Pouth[2][]）。
* 去实践一个学习时代。
*/
layerData = teacher.GetLayerInput(batches);
for (int i = 0; i < 200; i++)//200次学习
{
var learningResult = teacher.RunEpoch(layerData);
double error = learningResult / inputs.Length;//RunEpoch运行纪元 Returns sum of learning errors.
if (i % 10 == 0)
{
Console.WriteLine(i + ", Error = " + error);
}
}
}
// Supervised learning on entire network, to provide output classification.对整个网络进行监督学习，提供输出分类。
var teacher2 = new Neuro.Learning.BackPropagationLearning(network)
{
LearningRate = 0.1, //学习速率
Momentum = 0.5 //动力
};
// Run supervised learning.运行监督学习。
for (int i = 0; i < 500; i++)//500次学习
{
double error = teacher2.RunEpoch(inputs, outputs) / inputs.Length;
if (i % 10 == 0)
{
Console.WriteLine(i + ", Error = " + error);
}
}
// Test the resulting accuracy. 测试结果的准确性
int correct = 0;
for (int i = 0; i < inputs.Length; i++)
{
var cp = testInputs[i].ToList();
var cpn = testInputs[i + 1];
foreach (var item in cpn)
{
cp.Add(item);
}
double[] outputValues = network.Compute(cp.ToArray());
//double[] outputValues = network.Compute(testInputs[i]);
if (DataManager.FormatOutputResult(outputValues) == DataManager.FormatOutputResult(testOutputs[i]))
{
correct++;
}
}
Console.WriteLine("Correct " + correct + "/" + inputs.Length + ", " + System.Math.Round(((double)correct / (double)inputs.Length * 100), 2) + "%");
Console.Write("Press any key to quit ..");
Console.ReadKey();
}
