public void TrainNetworkByBatch(DataSamples samples, DataSamples testSamples, BackPropParams param)
{
var rnd = new Random();
//
infos.InitNeurons(rnd);
infos.InitNG(Network, rnd);
var stop = false;
var step = 0;
double error = 0.0;
var count = samples.Count;
while (!stop)
{
for (int sampleIndex = 0; sampleIndex < count; sampleIndex++)
{
var data = samples[sampleIndex];
// forward step
Network.Inputs = data.Inputs;
Network.Calc();
// backward
infos.ProcessGradients(data.Outputs /*, sampleIndex == 0*/);
infos.ProcessDeltas(param.Eta, param.Alpha, sampleIndex == 0);
}
infos.UpdateWeights(param.Alpha);
error = Network.Error(testSamples);
step++;
if (param.CallBack != null)
{
param.CallBack(step, error, false);
}
stop = (step >= param.MaxSteps) || (error <= param.ErrorStopValue);
}
if (param.CallBack != null)
{
param.CallBack(step, error, true);
}
}
public void TrainNetworkBySample(DataSamples samples, DataSamples testSamples, BackPropParams param)
{
var rnd = new Random();
var items = GenerateIndexes(samples);
//
infos.InitNeurons(rnd);
infos.InitNG(Network, rnd);
var stop = false;
var step = 0;
double error = 0.0;
while (!stop)
{
ShuffleIndexes(items, rnd);
for (int i = 0; i < items.Count; i++)
{
TrainBySample(samples[items[i]], param.Eta, param.Alpha);
}
//
error = Network.Error(testSamples);
step++;
if (param.CallBack != null)
{
param.CallBack(step, error, false);
}
stop = (step >= param.MaxSteps) || (error <= param.ErrorStopValue);
}
if (param.CallBack != null)
{
param.CallBack(step, error, true);
}
}