private void ShowResults()
{
if (rbBestNetwork.Checked)
{
mSelectedActivationNetwork = mBestActivationNetwork;
}
if (rbBestTestNetwork.Checked)
{
mSelectedActivationNetwork = mBestTestActivationNetwork;
}
if (rbGenetic.Checked)
{
mSelectedActivationNetwork = mBestGeneticActivationNetwork;
}
try
{
TVec dy = null;
TVec dx = null;
fittedChart.Series.Clear();
fittedChart.Series.Add(new Steema.TeeChart.Styles.FastLine());
fittedChart.Series.Add(new Steema.TeeChart.Styles.FastLine());
fittedChart.Series[0].Title = "наблюдения";
fittedChart.Series[1].Title = "прогнози";
fittedChart.Axes.Left.Title.Text = "Y";
fittedChart.Axes.Bottom.Title.Text = "";
fittedChart.Axes.Left.Automatic = true;
Vector timeseries = new Vector((TestEnd-ExtractBegin) + 1);
Vector abs = new Vector((TestEnd - ExtractBegin) + 1);
int? dependent_variable_id = 0;
foreach (int? dep in this.DependentVariables.Keys)
{
dependent_variable_id = dep;
}
MainDataSetTableAdapters.VARIABLESTableAdapter adptVariables = new Plig.TimeSeries.Client.MainDataSetTableAdapters.VARIABLESTableAdapter();
mindep = adptVariables.MinObservationValue(ExtractBegin, TestEnd, dependent_variable_id.Value).Value;
factordep = 1.7 / (adptVariables.MaxObservationValue(ExtractBegin, TestEnd, dependent_variable_id.Value).Value - adptVariables.MinObservationValue(ExtractBegin, TestEnd, dependent_variable_id.Value).Value);
int cnt = ExtractBegin;
foreach (KeyValuePair<int, double?> d in this.VariableObservations(dependent_variable_id.Value))
{
if (d.Key >= ExtractBegin && d.Key <= TestEnd)
{
try
{
timeseries.Values[cnt - ExtractBegin] = d.Value.Value;
}
catch
{
timeseries.Values[cnt - ExtractBegin] = 0.0;
}
abs.Values[cnt - ExtractBegin] = cnt;
cnt++;
}
}
dy = timeseries;
dx = abs;
Dew.Math.Tee.TeeChart.DrawValues(dx, dy, fittedChart.Series[0], false);
// Извличаме стойностите на независимите променливи
TVec dy1 = null;
Vector timeseries1 = new Vector(TestEnd - ExtractBegin + 1);
int independent_vars = this.IndependentVariables.Count;
double[] tuple = new double[independent_vars];
for (int i = ExtractBegin; i <= TestEnd; i++)
{
int cnt1 = 0;
foreach (int variable_id in this.IndependentVariables.Keys)
{
double factor = 1.7 / ( (this.SeriesInfo[variable_id]).Max - (this.SeriesInfo[variable_id]).Min);
tuple[cnt1] = (this.SeriesInfo[variable_id].Values[i].Value - (this.SeriesInfo[variable_id]).Min) * factor - 0.85;
cnt1++;
}
double value = (mSelectedActivationNetwork.Compute(tuple)[0] + 0.85) / factordep + mindep;
timeseries1.Values[i - ExtractBegin] = value;
}
dy1 = timeseries1;
Dew.Math.Tee.TeeChart.DrawValues(dx, dy1, fittedChart.Series[1], false);
residuals = new TVec();
residuals.Length = timeseries1.Length;
for (int i = 0; i < residuals.Length; i++)
{
residuals.Values[i] = timeseries1.Values[i] - timeseries.Values[i];
}
Vector Bins = new Vector(0);
Vector Freq = new Vector(0);
chartResiduals.Series[0].Title = "остатъци";
chartResiduals.Series[1].Title = "Нормално разпределение";
chartResiduals.Axes.Left.Title.Text = "Y";
chartResiduals.Axes.Bottom.Title.Text = "";
chartResiduals.Axes.Left.Automatic = true;
Statistics.Histogram(residuals, 20, Freq, Bins, true);
Dew.Math.Tee.TeeChart.DrawValues(Bins, Freq, chartResiduals.Series[0], false);
Vector xdense = new Vector(0);
Matrix tmpmtx = new Matrix(0, 0);
Vector x = new Vector(0);
Vector bell = new Vector(0);
// get values for Bell curve
double mean = residuals.Mean();
double stddev = residuals.StdDev(mean);
// 500 calc points
xdense.Length = 500;
xdense.Ramp(Bins.Min() , (Bins.Max() - Bins.Min()) * 0.005);
Dew.Math.Units.Probabilities.NormalPDF(xdense, mean, stddev, bell);
chartResiduals.Series[1].GetHorizAxis.Automatic = false;
chartResiduals.Series[1].GetHorizAxis.SetMinMax(residuals.Min(), residuals.Max());
Dew.Math.Tee.TeeChart.DrawValues(xdense, bell, chartResiduals.Series[1], false);
double R2 = CalculateR2(dy, dy1);
rchTests.Clear();
rchTests.SelectionColor = Color.Red;
rchTests.AppendText("Тестове");
rchTests.SelectionColor = SystemColors.InfoText; ;
rchTests.AppendText("\nНевронна мрежа с параметри: ");
rchTests.AppendText(String.Format("\nВходящ слой: {0} неврона", mSelectedActivationNetwork.InputsCount));
rchTests.AppendText(String.Format("\nСкрит слой: {0} неврона", mSelectedActivationNetwork[0].NeuronsCount));
rchTests.SelectionColor = Color.Red;
rchTests.AppendText("\nСтатистически параметри");
rchTests.SelectionColor = SystemColors.InfoText; ;
rchTests.AppendText(String.Format("\nCalculated R2: {0:F6}", R2));
rchTests.AppendText("\n\nАнализ на независимите променливи");
rchTests.AppendText("\nЗначимост");
double max_spread = 0.0;
foreach (int var_id in IndependentVariables.Keys)
{
Vector vec = IndepVar(var_id);
if (max_spread < vec.Max() - vec.Min())
{
max_spread = vec.Max() - vec.Min();
}
}
int current = 1;
foreach (int var_id in IndependentVariables.Keys)
{
Vector vec = IndepVar(var_id);
double spread = vec.Max() - vec.Min();
String result = String.Format("{0,-3}. {1,25}: {2,10} | {3}", current, IndependentVariables[var_id], spread, Stars( (int)Math.Floor((10*spread)/ max_spread)));
rchTests.AppendText("\n" + result);
current++;
}
rchTests.AppendText("\n\nDurbin-Watson тест за наличие на автоколинеарност в остатъците");
rchTests.AppendText(String.Format("\nDW: {0}", Utils.DurbinWatson((Vector)residuals)));
}
catch { };
}
private void btnResult_Click(object sender, EventArgs e)
{
try
{
TVec dy = null;
TVec dx = null;
fittedChart.Series.Clear();
fittedChart.Series.Add(new Steema.TeeChart.Styles.FastLine());
fittedChart.Series.Add(new Steema.TeeChart.Styles.FastLine());
fittedChart.Series[0].Title = "наблюдения";
fittedChart.Series[1].Title = "прогнози";
fittedChart.Axes.Left.Title.Text = "Y";
fittedChart.Axes.Bottom.Title.Text = "";
fittedChart.Axes.Left.Automatic = true;
Vector timeseries = new Vector((TestEnd - ExtractBegin) + 1);
Vector abs = new Vector((TestEnd - ExtractBegin) + 1);
int? dependent_variable_id = 0;
foreach (int? dep in this.DependentVariables.Keys)
{
dependent_variable_id = dep;
}
MainDataSetTableAdapters.VARIABLESTableAdapter adptVariables = new Plig.TimeSeries.Client.MainDataSetTableAdapters.VARIABLESTableAdapter();
double mindep = adptVariables.MinObservationValue(ExtractBegin, TestEnd, dependent_variable_id.Value).Value;
double factordep = 1.7 / (adptVariables.MaxObservationValue(ExtractBegin, TestEnd, dependent_variable_id.Value).Value - adptVariables.MinObservationValue(ExtractBegin, TestEnd, dependent_variable_id.Value).Value);
int cnt = ExtractBegin;
foreach (KeyValuePair<int, double?> d in this.VariableObservations(dependent_variable_id.Value))
{
if (d.Key >= ExtractBegin && d.Key <= TestEnd)
{
try
{
timeseries.Values[cnt - ExtractBegin] = d.Value.Value;
}
catch
{
timeseries.Values[cnt - ExtractBegin] = 0.0;
}
abs.Values[cnt - ExtractBegin] = cnt;
cnt++;
}
}
dy = timeseries;
dx = abs;
Dew.Math.Tee.TeeChart.DrawValues(dx, dy, fittedChart.Series[0], false);
// Извличаме стойностите на независимите променливи
TVec dy1 = null;
Vector timeseries1 = new Vector(TestEnd - ExtractBegin + 1);
int independent_vars = this.IndependentVariables.Count;
double[] tuple = new double[independent_vars];
Dew.Math.Tee.TeeChart.DrawValues(dx, multiLinReg.RegressResult.YCalc, fittedChart.Series[1], false);
}
catch (Exception ex)
{
}
}
protected override void StartTraining(bool aIsRunning)
{
if (aIsRunning)
{
mStopWatch = new Stopwatch();
mStopWatch.Start();
Stopwatch st = new Stopwatch();
st.Start();
min = Double.MaxValue;
mintest = Double.MaxValue;
// create grid application
Settings settings = Settings.LoadSettings();
GA = new GApplication(new GConnection(settings.AlchemiURL, settings.AlchemiPort, settings.AlchemiUser, settings.AlchemiPassword));
GA.ApplicationName = "Plain Neural Network model";
// add GridThread module (this executable) as a dependency
GA.Manifest.Add(new ModuleDependency(typeof(AForge.PolishExpression).Module));
GA.Manifest.Add(new ModuleDependency(typeof(AForge.Neuro.ActivationNetwork).Module));
GA.Manifest.Add(new ModuleDependency(typeof(SingleNNTraining).Module));
// subscribe to events
GA.ThreadFinish += new GThreadFinish(ThreadFinished);
GA.ThreadFailed += new GThreadFailed(ThreadFailed);
GA.ApplicationFinish += new GApplicationFinish(ApplicationFinished);
// Набор от данни използван за обучение
mInput = new double[ExtractEnd - ExtractBegin + 1][];
mOutput = new double[ExtractEnd - ExtractBegin + 1][];
// Тестов набор от данни
mTestInput = new double[TestEnd - TestBegin + 1][];
mTestOutput = new double[TestEnd - TestBegin + 1][];
for (int i = 0; i < ExtractEnd - ExtractBegin + 1; i++)
{
mInput[i] = new double[IndependentVariables.Count];
mOutput[i] = new double[DependentVariables.Count];
}
for (int i = 0; i < TestEnd - TestBegin + 1; i++)
{
mTestInput[i] = new double[IndependentVariables.Count];
mTestOutput[i] = new double[DependentVariables.Count];
}
MainDataSetTableAdapters.VARIABLESTableAdapter adptVariables = new Plig.TimeSeries.Client.MainDataSetTableAdapters.VARIABLESTableAdapter();
mReinitializations = Convert.ToInt32(txtReinitializations.Text);
results = new List<NNResultSet>(mReinitializations);
mEpochs = Convert.ToInt32(txtEpochs.Text);
ReinitsFinished = 0;
int index_of_variable = 0;
foreach (int variable_id in IndependentVariables.Keys)
{
double minObsValue = adptVariables.MinObservationValue(ExtractBegin, ExtractEnd, variable_id).Value;
double factor = 1.7 / (adptVariables.MaxObservationValue(ExtractBegin, ExtractEnd, variable_id).Value - adptVariables.MinObservationValue(ExtractBegin, ExtractEnd, variable_id).Value);
Dictionary<int, double?> observations = VariableObservations(variable_id);
foreach (int counter in observations.Keys)
{
if ((counter >= ExtractBegin) && (counter <= ExtractEnd))
{
mInput[counter - ExtractBegin][index_of_variable] = (observations[counter].Value - minObsValue) * factor - 0.85;
}
if ((counter >= TestBegin) && (counter <= TestEnd))
{
mTestInput[counter - TestBegin][index_of_variable] = (observations[counter].Value - minObsValue) * factor - 0.85;
}
}
index_of_variable++;
}
index_of_variable = 0;
foreach (int variable_id in DependentVariables.Keys)
{
double minObsValue = adptVariables.MinObservationValue(ExtractBegin, ExtractEnd, variable_id).Value;
double factor = 1.7 / (adptVariables.MaxObservationValue(ExtractBegin, ExtractEnd, variable_id).Value - adptVariables.MinObservationValue(ExtractBegin, ExtractEnd, variable_id).Value);
Dictionary<int, double?> observations = VariableObservations(variable_id);
foreach (int counter in observations.Keys)
{
if ((counter >= ExtractBegin) && (counter <= ExtractEnd))
{
mOutput[counter - ExtractBegin][index_of_variable] = (observations[counter].Value - minObsValue) * factor - 0.85;
}
if ((counter >= TestBegin) && (counter <= TestEnd))
{
mTestOutput[counter - TestBegin][index_of_variable] = (observations[counter].Value - minObsValue) * factor - 0.85;
}
}
index_of_variable++;
}
double momentum = 0.0;
double learning_rate = 0.1;
learning_rate = Convert.ToDouble(txtLearningRate.Text);
momentum = Convert.ToDouble(txtMomentum.Text);
int min_neurons = 1;
int max_neurons = 1;
if (rbFixedHiddenNeurons.Checked)
{
min_neurons = Convert.ToInt32(txtHiddenNeurons.Text);
max_neurons = min_neurons;
}
if (rbFlexibleHiddenNeurons.Checked)
{
min_neurons = Convert.ToInt32(txtMinHiddenNeurons.Text);
max_neurons = Convert.ToInt32(txtMaxHiddenNeurons.Text);
}
int TotalNNCalculations = 0;
for (int hidden_neurons = min_neurons; hidden_neurons <= max_neurons; hidden_neurons++)
{
for (int i = 0; i < mReinitializations; i++)
// Parallel.For(0, mReinitializations - 1, i =>
{
// create thread
SingleNNTraining single_nn = new SingleNNTraining(mInput, mOutput, mEpochs);
single_nn.HiddenNeurons = hidden_neurons;
single_nn.Iterations = mEpochs;
single_nn.Momentum = momentum;
single_nn.LearningRate = learning_rate;
// добавяне на набора от данни за тестване
single_nn.TestInput = mTestInput;
single_nn.TestOutput = mTestOutput;
// add thread to application
GA.Threads.Add(single_nn);
TotalNNCalculations++;
}
// );
}
pbProgressTraining.Minimum = 0;
pbProgressTraining.Maximum = TotalNNCalculations - 1;
GA.Start();
st.Stop();
lbResults.Items.Add("Време необходимо за конструиране на данните: " + st.Elapsed.ToString());
int populationCount = Convert.ToInt32(txtPopulationCount.Text);
Population population = new Population(populationCount,
new DoubleArrayChromosome(new AForge.Math.Random.UniformOneGenerator(),
new AForge.Math.Random.GaussianGenerator(0, 1),new AForge.Math.Random.GaussianGenerator(0, 1),
(mOutput.Length + mInput.Length + 1) * 10 + mOutput.Length),
new NeuralNetworkFitness(mInput, mOutput, new int[] { IndependentVariables.Count, 10, DependentVariables.Count }),
new EliteSelection());
int geneticEpochs = Convert.ToInt32(txtGeneticEpochs.Text);
pbGeneticProgress.Maximum = geneticEpochs;
for (int i = 0; i < geneticEpochs; i++)
{
population.RunEpoch();
pbGeneticProgress.Value = i + 1;
Application.DoEvents();
}
NeuralNetworkFitness nnf = new NeuralNetworkFitness(mInput, mOutput, new int[] { IndependentVariables.Count, 5, DependentVariables.Count });
mBestGeneticActivationNetwork = (ActivationNetwork)nnf.Translate(population.BestChromosome);
}
else
{
if (GA.Running)
{
foreach (GThread thread in GA.Threads)
{
thread.Abort();
}
// GA.Stop();
lbResults.Items.Add("Обучението е прекъснато.");
}
}
}