public void TestPredictionWindowSize()
{
using (StreamWriter sw = new StreamWriter(@"E:\repos\RBFNetwork\windowSize.txt"))
{
for (int i = 10; i <= 10; i++)
{
var data = GetTemperatureDataAsync(95, i).Result;
int accuracy = 0;
var topology = new Topology(data.trainData.ClassesNames?.Count ?? 1, 0.01, data.trainData.ColumnCount);
var network = new RBFNeuralNetwork(topology);
double error = network.Train(data.trainData, 100, false);
double predictError = 0;
foreach (Record record in data.testData.Records)
{
var expected = record.Expected[0];
var output = network.Predict(record.InputData)[0];
var diff = expected - output;
predictError += diff * diff;
if ((expected * data.inputsSum - output * data.inputsSum) <= 2)
{
accuracy++;
}
}
predictError = Math.Sqrt(predictError / (data.testData.RowCount - 1));
Trace.WriteLine($"Current accuracy: {accuracy} of {data.testData.Records.Count} = {accuracy * 100 / data.testData.Records.Count}%");
var predicted = accuracy * 100 / data.testData.RowCount;
sw.WriteLine($"{i} {predictError} {predicted}");
}
}
}
// Прогнозирование
public List <DataPoint> Predict()
{
List <DataPoint> result = new List <DataPoint>();
int k = 1;
// Для нормализации
double inputsSum = testDataset.Records.Sum(r => r.InputData.Sum(x => x * x) + r.Expected.Sum(x => x * x));
inputsSum = Math.Sqrt(inputsSum);
// Нормализация
testDataset.Normalize();
foreach (Record record in testDataset.Records)
{
var expected = record.Expected[0];
var output = neuralNetwork.Predict(record.InputData)[0];
// Разница между ожидаемым и выходным
var diff = expected - output;
// Набор точек на графике
// к - точка на графике
var dp = new DataPoint(expected * inputsSum, output * inputsSum, diff * diff, k);
result.Add(dp);
k++;
}
return(result);
}
