void drawCelsiusFaringate2()
{
var ira = new IRINA_NN(new int[] { 1, 6, 9, 3, 1 }, 0.2);
var input = new List <double[]>
{
new double[] { 0.040 },
new double[] { 0.035 },
new double[] { 0.037 },
new double[] { 0.048 },
new double[] { -0.012 },
new double[] { -0.040 },
new double[] { 0.057 },
new double[] { 0.089 },
new double[] { -0.005 },
new double[] { 0.100 },
new double[] { -0.042 }
};
var target = new List <double[]>
{
new double[] { 0.104 },
new double[] { 0.095 },
new double[] { 0.0986 },
new double[] { 0.1184 },
new double[] { 0.0104 },
new double[] { -0.040 },
new double[] { 0.1346 },
new double[] { 0.1922 },
new double[] { 0.023 },
new double[] { 0.212 },
new double[] { -0.0436 }
};
var samplesCount = input.Count;
var random = new Random();
for (int epoch = 0; epoch < count; ++epoch)
{
var index = random.Next(0, samplesCount);
var inputItem = input[index];
var targetItem = target[index];
ira.Train(inputItem, targetItem);
var error = IRINA_NN.MSE(ira.Predict(inputItem), inputItem);
if (epoch % even == 0)
{
SeriesCollection[0].Values.Add(error);
}
}
var result = ira.Predict(new double[] { 0.67 })[0];
result = 0d;
}
void drawXor()
{
var ira = new IRINA_NN(new int[] { 2, 4, 4, 1 }, 0.2);
var e1 = ira.Predict(input: new double[] { 0.99, 0.99 })[0];
var e2 = ira.Predict(input: new double[] { 0.99, 0.01 })[0];
var e3 = ira.Predict(input: new double[] { 0.01, 0.99 })[0];
var e4 = ira.Predict(input: new double[] { 0.01, 0.01 })[0];
var error = (e1 + e2 + e3 + e4) / 4d;
for (int epoch = 0; epoch < count; ++epoch)
{
ira.Train(
input: new double[] { 0.01, 0.01 },
target: new double[] { 0.01 }
);
e1 = IRINA_NN.MSE(ira.Predict(new double[] { 0.01, 0.01 }), new double[] { 0.01 });
ira.Train(
input: new double[] { 0.99, 0.01 },
target: new double[] { 0.99 }
);
e2 = IRINA_NN.MSE(ira.Predict(new double[] { 0.99, 0.01 }), new double[] { 0.99 });
ira.Train(
input: new double[] { 0.01, 0.99 },
target: new double[] { 0.99 }
);
e3 = IRINA_NN.MSE(ira.Predict(new double[] { 0.01, 0.99 }), new double[] { 0.99 });
ira.Train(
input: new double[] { 0.99, 0.99 },
target: new double[] { 0.01 }
);
e4 = IRINA_NN.MSE(ira.Predict(new double[] { 0.99, 0.99 }), new double[] { 0.01 });
error = (e1 + e2 + e3 + e4) / 4d;
//lbOutput.Items.Add($"Ошибка эпохи { epoch }: { error }");
if (epoch % even == 0)
{
SeriesCollection[0].Values.Add(error);
}
}
e2 = ira.Predict(input: new double[] { 0.99, 0.01 })[0];
e1 = ira.Predict(input: new double[] { 0.99, 0.99 })[0];
e4 = ira.Predict(input: new double[] { 0.01, 0.01 })[0];
e3 = ira.Predict(input: new double[] { 0.01, 0.99 })[0];
e1 = e2 = e3 = e4;
}
void drawCelsiusFaringate()
{
var ira = new IRINA_NN(new int[] { 8, 16, 32, 16, 8 }, 0.2);
for (int epoch = 0; epoch < count; ++epoch)
{
ira.Train( // 40 - 104
input: new double[] { 0.01, 0.01, 0.99, 0.01, 0.99, 0.01, 0.01, 0.01 },
target: new double[] { 0.01, 0.99, 0.99, 0.01, 0.99, 0.01, 0.01, 0.01 }
);
var e1 = IRINA_NN.MSE(ira.Predict(new double[] { 0.01, 0.01, 0.99, 0.01, 0.99, 0.01, 0.01, 0.01 }), new double[] { 0.01, 0.99, 0.99, 0.01, 0.99, 0.01, 0.01, 0.01 });
ira.Train( // 60 - 140
input: new double[] { 0.01, 0.01, 0.99, 0.99, 0.99, 0.99, 0.01, 0.01 },
target: new double[] { 0.99, 0.01, 0.01, 0.01, 0.01, 0.99, 0.99, 0.01 }
);
var e2 = IRINA_NN.MSE(ira.Predict(new double[] { 0.01, 0.01, 0.99, 0.99, 0.99, 0.99, 0.01, 0.01 }), new double[] { 0.99, 0.01, 0.01, 0.01, 0.01, 0.99, 0.99, 0.01 });
ira.Train( // 15 - 59
input: new double[] { 0.01, 0.01, 0.01, 0.01, 0.99, 0.99, 0.99, 0.99 },
target: new double[] { 0.01, 0.01, 0.99, 0.99, 0.99, 0.01, 0.99, 0.99 }
);
var e3 = IRINA_NN.MSE(ira.Predict(new double[] { 0.01, 0.01, 0.01, 0.01, 0.99, 0.99, 0.99, 0.99 }), new double[] { 0.01, 0.01, 0.99, 0.99, 0.99, 0.01, 0.99, 0.99 });
ira.Train( // 25 - 77
input: new double[] { 0.01, 0.01, 0.01, 0.99, 0.99, 0.01, 0.01, 0.99 },
target: new double[] { 0.01, 0.99, 0.01, 0.01, 0.99, 0.99, 0.01, 0.99 }
);
var e4 = IRINA_NN.MSE(ira.Predict(new double[] { 0.01, 0.01, 0.01, 0.99, 0.99, 0.01, 0.01, 0.99 }), new double[] { 0.01, 0.99, 0.01, 0.01, 0.99, 0.99, 0.01, 0.99 });
var error = (e1 + e2 + e3 + e4) / 4d;
if (epoch % even == 0)
{
SeriesCollection[0].Values.Add(error);
}
}
var predicted = ira.Predict(new double[] { 0.01, 0.01, 0.01, 0.99, 0.99, 0.01, 0.01, 0.99 });
var target = new double[] { 0.01, 0.99, 0.01, 0.01, 0.99, 0.99, 0.01, 0.99 };
var err = IRINA_NN.MSE(predicted, target);
predicted = ira.Predict(new double[] { 0.01, 0.01, 0.99, 0.01, 0.01, 0.01, 0.99, 0.99 });
target = new double[] { 0.01, 0.99, 0.01, 0.99, 0.99, 0.99, 0.99, 0.99 };
err = IRINA_NN.MSE(predicted, target);
}
public static void Export(IRINA_NN network, string fileName)
{
try
{
using (var fileStream = new FileStream(fileName, FileMode.Create))
{
BinaryFormatter formatter = new BinaryFormatter();
formatter.Serialize(fileStream, network);
}
}
catch (Exception e)
{
throw new Exception("Ошибка экспорта", e);
}
}
void drawIris()
{
var ira = new IRINA_NN(new int[] { 4, 16, 64, 24, 3 }, 0.02);
IrisDataReader irisDataReader = new IrisDataReader(@"D:\NeuralNetworks\DataSets\Iris Plants Database\bezdekIris.data");
var rand = new Random();
for (int epoch = 0; epoch < count; ++epoch)
{
var irisData = irisDataReader.Read(rand.Next(0, 150));
var(input, target) = irisToNN(irisData);
ira.Train(input, target);
var predict = ira.Predict(input);
var error = IRINA_NN.MSE(predict, target);
if (epoch % even == 0)
{
SeriesCollection[0].Values.Add(error);
}
}
var data = irisDataReader.Read(25);
var(inputRes, targetRes) = irisToNN(data);
var predictRes = ira.Predict(inputRes);
var errorRes = IRINA_NN.MSE(predictRes, targetRes);
data = irisDataReader.Read(72);
(inputRes, targetRes) = irisToNN(data);
predictRes = ira.Predict(inputRes);
errorRes = IRINA_NN.MSE(predictRes, targetRes);
data = irisDataReader.Read(119);
(inputRes, targetRes) = irisToNN(data);
predictRes = ira.Predict(inputRes);
errorRes = IRINA_NN.MSE(predictRes, targetRes);
}
private void Window_Loaded(object sender, RoutedEventArgs e)
{
ira = IRINA_NN.Import(@"mnist.nn");
reader = new MNISTDataReader.ImageDataReader(@"D:\NeuralNetworks\DataSets\MNIST\t10k-images.idx3-ubyte");
lablesReader = new MNISTDataReader.LabelDataReader(@"D:\NeuralNetworks\DataSets\MNIST\t10k-labels.idx1-ubyte");
Predict();
//IrisDataReader irisDataReader = new IrisDataReader(@"D:\NeuralNetworks\DataSets\Iris Plants Database\bezdekIris.data");
//var rand = new Random();
//while(true)
//{
// irisDataReader.Read(rand.Next(0, 150));
//}
//calcError();
}
void drawMNIST()
{
var ira = new IRINA_NN(new int[] { 784, 200, 10 }, 0.2);
var imageReader = new MNISTDataReader.ImageDataReader(@"D:\NeuralNetworks\DataSets\MNIST\train-images.idx3-ubyte");
var labelReader = new MNISTDataReader.LabelDataReader(@"D:\NeuralNetworks\DataSets\MNIST\train-labels.idx1-ubyte");
var itemsCount = imageReader.ItemsCount;
var random = new Random();
Stopwatch stopwatch = new Stopwatch();
stopwatch.Start();
for (int epoch = 0; epoch < count; ++epoch)
{
//var errors = new List<double>();
for (int sample = 0; sample < samplesInBatch; ++sample)
{
var realIndex = random.Next(0, itemsCount);
var input = prepareInput(imageReader.Read(realIndex));
var label = labelReader.Read(realIndex);
var target = prepareTarget(label);
ira.Train(input, target);
//var predicted = ira.Predict(input);
//errors.Add(0.99 - predicted[label]);
}
//SeriesCollection[0].Values.Add(errors.Average());
}
stopwatch.Stop();
IRINA_NN.Export(ira, @"mnist.nn");
MessageBox.Show($"Complete { count } epochs, { samplesInBatch } samples per one. Estimated time: { stopwatch.ElapsedMilliseconds }ms");
}