我有一个关于如何在encog库c#中保存神经网络的问题。我想从输入和输出层序列化隐藏层和数据的权重。如果我想成功反序列化,还有必要保存网络的某个结构。 在我创建网络和序列化BasicNetwork对象的代码部分下面,当然它是不正确的。我已经找到了很多关于如何使用java版本的信息,但注意到了c#。
BasicNetwork network = CreateNet(nettype,res11[i],1,2);
INeuralDataSet trainingSet = new BasicNeuralDataSet(masStudyInput, masStudyOutput);
INeuralDataSet TestingSet = new BasicNeuralDataSet(masTestInput, mastestOutput);
ITrain train = new ResilientPropagation(network, trainingSet);
int epoch = 1;
//network.Structure.Layers.
MessageBox.Show("Start");
do
{
train.Iteration();
mist = GetMistake(ref network, ref TestingSet);
chart1.Invoke((Action)(() =>
{
chart1.Series[0].Points.AddY(train.Error);
chart1.Series[1].Points.AddY(mist);
}));
network.
if (mist < 0.8)
{
string XMLfilename = System.IO.Path.GetDirectoryName(Application.ExecutablePath) + @"\" + mist + ".xml";
System.Xml.Serialization.XmlSerializer writer = new System.Xml.Serialization.XmlSerializer(typeof(BasicNetwork));
TextWriter writerr = new StreamWriter(XMLfilename);
writer.Serialize(writerr, network);
writerr.Close();
}
epoch++;
}
while ((epoch < 1000));SS
答案 0 :(得分:8)
Encog还有saver / loader:
EncogDirectoryPersistence.SaveObject(new System.IO.FileInfo(txtSaveFile.Text), _network);
_network = (BasicNetwork)EncogDirectoryPersistence.LoadObject(new FileInfo(txtSaveFile.Text));
答案 1 :(得分:2)
网络是可序列化的,因此您可以使用任何所需的序列化程序。 一个例子:
var serializer = new BinaryFormatter();
using (var ms = new MemoryStream())
{
serializer.Serialize(ms, network);
ms.Position = 0;
byte[] serialized = ms.ToArray();
return serialized;
}