private void teachButton_Click(object sender, EventArgs e)
{
_terminate = false;
_error = 1.0;
_progress = 0;
_terminated = false;
_controlError = -1;
_updateProgress();
new Thread(() =>
{
var samples = CaptchasLibrary.GetRecognizedCaptchas();
var digitImages = new List <Image>();
var digitResults = new List <int>();
int controlStart = (int)(samples.Count * 0.8);
for (var i = 0; i < controlStart; i++)
{
var recognizer = new CaptchaRecognizer(samples[i].Image);
var rectangles = recognizer.SplitDigits();
for (var j = 0; j < rectangles.Length; j++)
{
digitImages.Add(ImageHelper.Crop(recognizer.Target, rectangles[j]));
digitResults.Add(int.Parse(samples[i].Answer.Substring(j, 1)));
}
}
var teacher = new NeuralNetworkTrainer(digitImages.ToArray(), digitResults.ToArray());
teacher.Run((net, data, epochs, reports, er, u, userData) =>
{
_error = net.MSE;
_progress = 1.0 * u / epochs;
if (_terminate)
{
return(-1);
}
return(0);
});
teacher.Network.Save(CaptchaRecognizer.NetworkPath);
var controlTotal = 0;
var controlPasses = 0;
for (var i = controlStart; i < samples.Count; i++)
{
var rec = new CaptchaRecognizer(samples[i].Image);
var res = rec.Recognize();
controlTotal++;
if (res.Result == samples[i].Answer)
{
controlPasses++;
}
}
_controlError = 1.0 * controlPasses / controlTotal;
_error = teacher.Network.MSE;
_terminate = true;
_terminated = true;
_progress = 1;
}).Start();
}
private double[] _computeProbabs(Image image)
{
return(_network.Run(NeuralNetworkTrainer.GetRawData(image)));
}
