public NeuralTrainResult SimpleTrain()
{
if (DataSets == null || Net == null)
{
return(null);
}
Net.Randomize();
var result = new NeuralTrainResult
{
EpochCount = EpochsCount,
ErrorWriteCycle = ErrorWriteCycle,
Start = DateTime.UtcNow,
TolerableError = TolerableError,
TrainSets = DataSets,
TargetNet = Net,
ErrorHistory = new List <double>(),
EpochFinished = 0,
LearnRate = LearnRate
};
var counter = EpochsCount;
var totalError = 0d;
for (; counter > 0; counter--)
{
result.EpochFinished++;
totalError = 0d;
for (var i = 0; i < DataSets.Length; i++)
{
var input = DataSets[i].Item1;
var correctOut = DataSets[i].Item2;
var currentOut = Net.Activate(input);
var outDelta = new double[currentOut.Length];
for (var j = 0; j < outDelta.Length; j++)
{
outDelta[j] = correctOut[j] - currentOut[j];
totalError += Math.Pow(outDelta[j], 2);
}
Net.Adjust(correctOut, LearnRate);
}
totalError /= DataSets.Length;
if (Log != null && counter % ErrorWriteCycle == 0)
{
result.ErrorHistory.Add(totalError);
Log(EpochsCount, EpochsCount - counter, TolerableError, LearnRate, totalError);
}
if (totalError <= TolerableError)
{
break;
}
}
result.ResultError = totalError;
result.Stop = DateTime.UtcNow;
return(result);
}