public bool Training_Loop(Regression training_data, int step_size = 1, int batch_size = 128)
{
if (step_size <= 0 || batch_size <= 0)
{
return(false);
}
if (batch_size > training_data.Length)
{
batch_size = training_data.Length;
}
for (int s = 0; s < step_size; s++)
{
if (!infinite_loop)
{
if (circle >= max_circle)
{
return(false);
}
}
if (!ignore_error)
{
if (total_error <= error_limit)
{
return(false);
}
}
total_error = 0f;
var batch_counter = 0;
var batch_limit = batch_size;
while (batch_counter < training_data.Length)
{
var batch_regression = new Regression();
for (var i = batch_counter; i < batch_limit && i < training_data.Length; i++)
{
batch_regression.Add(training_data.inputs[i], training_data.outputs[i]);
}
var len = batch_regression.Length;
for (var i = 0; i < len; i++)
{
Training(batch_regression.inputs[i], batch_regression.outputs[i]);
total_error += GetError();
}
batch_counter += batch_size;
batch_limit += batch_size;
}
circle++;
}
return(true);
}
private void LoadFittedCurve()
{
if (Points.Count > 0 && _fittedCurve != null)
{
Regression.Clear();
var result = _fittedCurve.CalcFit(Points.ToList());
result.Points.ForEach(x => Regression.Add(x));
Title = result.Title;
SubTitle = $"A : {result.A} , B : {result.B} , R : {result.R}";
}
}
public void Training(Regression training_data, int batch_size = 128)
{
if (batch_size <= 0)
{
return;
}
if (batch_size > training_data.Length)
{
batch_size = training_data.Length;
}
circle = 0;
do
{
total_error = 0f;
var batch_counter = 0;
var batch_limit = batch_size;
while (batch_counter < training_data.Length)
{
var batch_regression = new Regression();
for (var i = batch_counter; i < batch_limit && i < training_data.Length; i++)
{
batch_regression.Add(training_data.inputs[i], training_data.outputs[i]);
}
var len = batch_regression.Length;
for (var i = 0; i < len; i++)
{
Training(batch_regression.inputs[i], batch_regression.outputs[i]);
total_error += GetError();
}
batch_counter += batch_size;
batch_limit += batch_size;
batch_regression.Dispose();
}
if (!infinite_loop)
{
if (circle >= max_circle)
{
break;
}
}
circle++;
} while (!ignore_error && (total_error >= error_limit));
}
public void Training(Classification trainingData)
{
Regression regressionTrainingData = new Regression();
for (int i = 0; i < trainingData.Length; i++)
{
float[] outputs = new float[output_layer.Length];
for (int j = 0; j < outputs.Length; j++)
{
if (trainingData.tags[i].Equals(output_layer[j].tag))
{
outputs[j] = 1f;
}
else
{
outputs[j] = 0f;
}
}
regressionTrainingData.Add(trainingData.inputs[i], outputs);
}
Training(regressionTrainingData);
}
private void PictureBoxPoster_MouseUp(object sender, MouseEventArgs e)
{
toggleMove = false;
PictureBox pictureBox = (PictureBox)sender;
bool train = false;
float value = 0f;
if (pictureBox.Location.X >= (this.Width - (this.Width * padding)))
{
value = 1f;
train = true;
existRight = true;
}
else if (pictureBox.Location.X <= (this.Width * padding - size.Width))
{
value = 0f;
train = true;
existLeft = true;
}
else if (trainingData.Length > 0 && (existLeft && existRight))
{
pictureBox.Location = new Point(lastPoint.X, pictureBox.Location.Y);
pictureBox.BringToFront();
}
if (train)
{
string[] split = linesList[int.Parse(pictureBox.Tag.ToString())].Split(' ');
int day = int.Parse(split[1]) - 1;
int month = months.IndexOf(split[2]);
float year = Map.Float(0f, float.Parse(split[3]), 9999f, 0f, 1f);
List <float> inputs = new List <float>();
for (int i = 0; i < 31; i++)
{
if (i == day)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
for (int i = 0; i < 12; i++)
{
if (i == month)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
inputs.Add(year);
string[] categoriesArr = split[4].Split(',');
for (int i = 0; i < categories.Count; i++)
{
bool exist = false;
for (int j = 0; j < categoriesArr.Length; j++)
{
if (categories[i] == categoriesArr[j])
{
exist = true;
break;
}
}
if (exist)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
string[] directorsArr = split[5].Split(',');
for (int i = 0; i < directors.Count; i++)
{
bool exist = false;
for (int j = 0; j < directorsArr.Length; j++)
{
if (directors[i] == directorsArr[j])
{
exist = true;
break;
}
}
if (exist)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
string[] starsArr = split[6].Split(',');
for (int i = 0; i < stars.Count; i++)
{
bool exist = false;
for (int j = 0; j < starsArr.Length; j++)
{
if (stars[i] == starsArr[j])
{
exist = true;
break;
}
}
if (exist)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
trainingData.Add(inputs.ToArray(), new[] { value });
stopwatch.Start();
neuralNetwork.Training(trainingData);
stopwatch.Stop();
Console.WriteLine("Training : " + stopwatch.ElapsedMilliseconds);
stopwatch.Reset();
pictureBox.MouseDown -= PictureBoxPoster_MouseDown;
pictureBox.MouseUp -= PictureBoxPoster_MouseUp;
pictureBox.MouseMove -= PictureBoxPoster_MouseMove;
pictureBoxes.Remove(pictureBox);
if (existLeft && existRight)
{
for (int i = 0; i < pictureBoxes.Count; i++)
{
split = linesList[int.Parse(pictureBoxes[i].Tag.ToString())].Split(' ');
day = int.Parse(split[1]) - 1;
month = months.IndexOf(split[2]);
year = Map.Float(1f, float.Parse(split[3]), 9999f, 0f, 1f);
inputs = new List <float>();
for (int j = 0; j < 31; j++)
{
if (j == day)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
for (int j = 0; j < 12; j++)
{
if (j == month)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
inputs.Add(year);
categoriesArr = split[4].Split(',');
for (int j = 0; j < categories.Count; j++)
{
bool exist = false;
for (int z = 0; z < categoriesArr.Length; z++)
{
if (categories[j] == categoriesArr[z])
{
exist = true;
break;
}
}
if (exist)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
directorsArr = split[5].Split(',');
for (int j = 0; j < directors.Count; j++)
{
bool exist = false;
for (int z = 0; z < directorsArr.Length; z++)
{
if (directors[j] == directorsArr[z])
{
exist = true;
break;
}
}
if (exist)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
starsArr = split[6].Split(',');
for (int j = 0; j < stars.Count; j++)
{
bool exist = false;
for (int z = 0; z < starsArr.Length; z++)
{
if (stars[j] == starsArr[z])
{
exist = true;
break;
}
}
if (exist)
{
inputs.Add(1f);
}
else
{
inputs.Add(0f);
}
}
stopwatch.Start();
Result[] guesses = neuralNetwork.Predict(inputs.ToArray());
stopwatch.Stop();
Console.WriteLine("Predict:" + stopwatch.ElapsedMilliseconds);
stopwatch.Reset();
pictureBoxes[i].Location = new Point((int)Map.Float(0f,
guesses[0].value,
1f,
this.Width * padding,
this.Width - (this.Width * padding) - size.Width),
(int)RandomF.NextFloat(this.Height - size.Height));
}
}
}
}