public static void test()
{
AvilaDataProcessing dataProcessing = new AvilaDataProcessing(new AvilaDataConverter(new AvilaClassifier()));
AvilaEvaluater evaluater = new AvilaEvaluater(new AvilaDataConverter(new AvilaClassifier()));
var inputDataSet = dataProcessing.GetData(@"C:\Work\DataSets\avila\avila-tr.txt");
var processDataSet = dataProcessing.GetProcessDataCollection(10000, inputDataSet);
ProcessData[] trainDataSet;
ProcessData[] testDataSet;
(trainDataSet, testDataSet) = dataProcessing.DivideCollection(0.7, processDataSet);
trainDataSet = trainDataSet.Take(6000).ToArray();
//Model avilaFNN = new Model(new CrossEntropyLoss(),
// new FullConnectLayer(new Direct(), 10, 10, true, new Normal()),
// new FullConnectLayer(new Sigmoid(), 10, 12, false, new Normal()),
// new FullConnectLayer(new LeakyReLU(0.1), 12, 16, false, new Normal()),
// new SoftMaxLayer(16, 12, new Normal())
// );
Model avilaFNN = new Model(new CrossEntropyLoss(), "AvilaFNN",
new FullConnectLayer(new Direct(1), 10, 10, true, new Normal()),
new FullConnectLayer(new Sigmoid(), 10, 16, false, new Normal(0, 2)),
new SoftMaxLayer(16, 12, new Normal(0, 5))
);
Trainer trainer1 = new Trainer(100, false, 0, new ExponentialDelayOptimizer(1, 0.9, 0.1));
//Trainer trainer2 = new Trainer(200, false, 0, new ExponentialDelayOptimizer(2, 0.95, 0.08));
avilaFNN = (Model)trainer1.Train(avilaFNN, trainDataSet);
trainDataSet = trainDataSet.SelectPermutation().ToArray();
//Console.WriteLine(avilaFNN.layers[2].Weight[0,0]);
Console.WriteLine(evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
avilaFNN = (Model)trainer1.Train(avilaFNN, trainDataSet);
//Console.WriteLine(avilaFNN.layers[2].Weight[0, 0]);
Console.WriteLine(evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
//for (int i = 0; i <= 49; i++)
//{
// trainDataSet = trainDataSet.SelectPermutation().ToList();
// Trainer trainer2 = new Trainer(60, false, 0, new ExponentialDelayOptimizer(2, 0.95, 0.05));
// avilaFNN = (Model)trainer2.Train(avilaFNN, trainDataSet);
// //Console.WriteLine(avilaFNN.layers[2].Weight[0, 0]);
// Console.WriteLine($"第1轮第{i}次循环:"+evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
//}
//for (int i = 0; i <= 49; i++)
//{
// trainDataSet = trainDataSet.SelectPermutation().ToList();
// Trainer trainer2 = new Trainer(200, false, 0, new ExponentialDelayOptimizer(5, 0.9, 0.05));
// avilaFNN = (Model)trainer2.Train(avilaFNN, trainDataSet);
// //Console.WriteLine(avilaFNN.layers[2].Weight[0, 0]);
// Console.WriteLine($"第2轮第{i}次循环:"+ evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
//}
//for (int i = 0; i <= 1; i++)
//{
// trainDataSet = trainDataSet.SelectPermutation().ToList();
// Trainer trainer2 = new Trainer(600, false, 0, new ExponentialDelayOptimizer(10, 0.9, 0.02));
// avilaFNN = (Model)trainer2.Train(avilaFNN, trainDataSet);
// //Console.WriteLine(avilaFNN.layers[2].Weight[0, 0]);
// Console.WriteLine(evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
//}
//for (int i = 0; i <= 99; i++)
//{
// trainDataSet = trainDataSet.SelectPermutation().ToList();
// Trainer trainer2 = new Trainer(200, false, 0, new ExponentialDelayOptimizer(5, 0.95, 0.02));
// avilaFNN = (Model)trainer2.Train(avilaFNN, trainDataSet);
// //Console.WriteLine(avilaFNN.layers[2].Weight[0, 0]);
// Console.WriteLine($"第3轮第{i}次循环:"+evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
//}
//for (int i = 0; i <= 1; i++)
//{
// trainDataSet = trainDataSet.SelectPermutation().ToList();
// Trainer trainer2 = new Trainer(600, false, 0, new ExponentialDelayOptimizer(10, 0.9, 0.02));
// avilaFNN = (Model)trainer2.Train(avilaFNN, trainDataSet);
// //Console.WriteLine(avilaFNN.layers[2].Weight[0, 0]);
// Console.WriteLine(evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
//}
//for (int i = 0; i <= 149; i++)
//{
// trainDataSet = trainDataSet.SelectPermutation().ToList();
// Trainer trainer2 = new Trainer(100, false, 0, new ExponentialDelayOptimizer(1, 0.98, 0.01));
// avilaFNN = (Model)trainer2.Train(avilaFNN, trainDataSet);
// //Console.WriteLine(avilaFNN.layers[2].Weight[0, 0]);
// Console.WriteLine($"第4轮第{i}次循环:"+evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
//}
evaluater.EvaluateInDetails(testDataSet, avilaFNN, new AvilaClassifier());
}
public static void train(string dataPath)
{
//define dataprocesser
AvilaDataProcessing dataProcessing = new AvilaDataProcessing(new AvilaDataConverter(new AvilaClassifier()));
//define evaluater
AvilaEvaluater evaluater = new AvilaEvaluater(new AvilaDataConverter(new AvilaClassifier()));
//create dataset
var inputDataSet = dataProcessing.GetData(dataPath);
var processDataSet = dataProcessing.GetProcessDataCollection(10000, inputDataSet);
ProcessData[] trainDataSet;
ProcessData[] testDataSet;
(trainDataSet, testDataSet) = dataProcessing.DivideCollection(0.7, processDataSet);
trainDataSet = trainDataSet.Take(6000).ToArray();
//define the model
Model avilaFNN = new Model(new CrossEntropyLoss(), "AvilaFNN",
new FullConnectLayer(new Direct(1), 10, 10, true, new Normal()),
new FullConnectLayer(new Sigmoid(), 10, 16, false, new Normal()),
new SoftMaxLayer(16, 12, new Normal())
);
//define optimizer
Trainer trainer = new Trainer(100, false, 0, new ExponentialDelayOptimizer(2, 0.95, 0.1));
//train
for (int i = 0; i <= 19; i++)
{
trainDataSet = trainDataSet.SelectPermutation().ToArray();
trainer = new Trainer(100, false, 0, new ExponentialDelayOptimizer(0.5, 0.9, 0.08));
avilaFNN = (Model)trainer.Train(avilaFNN, trainDataSet);
Console.WriteLine($"epoch {i}:" + evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
}
for (int i = 0; i <= 19; i++)
{
trainDataSet = trainDataSet.SelectPermutation().ToArray();
trainer = new Trainer(60, false, 0, new ExponentialDelayOptimizer(0.5, 0.9, 0.05));
avilaFNN = (Model)trainer.Train(avilaFNN, trainDataSet);
Console.WriteLine($"epoch {i}:" + evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
}
for (int i = 0; i <= 19; i++)
{
trainDataSet = trainDataSet.SelectPermutation().ToArray();
trainer = new Trainer(30, false, 0, new ExponentialDelayOptimizer(0.2, 0.9, 0.05));
avilaFNN = (Model)trainer.Train(avilaFNN, trainDataSet);
Console.WriteLine($"epoch {i}:" + evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
}
//for (int i = 0; i <= 49; i++)
//{
// trainDataSet = trainDataSet.SelectPermutation().ToArray();
// trainer = new Trainer(5, false, 0, new ExponentialDelayOptimizer(0.05, 0.9, 0.1));
// avilaFNN = (Model)trainer.Train(avilaFNN, trainDataSet);
// Console.WriteLine($"epoch {i}:" + evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
//}
//for (int i = 0; i <= 99; i++)
//{
// trainer = new Trainer(64, false, 0, new ExponentialDelayOptimizer(0.2, 0.95, 0.1));
// avilaFNN = (Model)trainer.Train(avilaFNN, trainDataSet);
// Console.WriteLine($"epoch {i}:" + evaluater.Evaluate(testDataSet, avilaFNN, new AvilaClassifier()));
//}
//evaluate
evaluater.EvaluateInDetails(trainDataSet, avilaFNN, new AvilaClassifier());
evaluater.EvaluateInDetails(testDataSet, avilaFNN, new AvilaClassifier());
}
