private TuningParamsClass step_gradient(double b_current, double m_current, List <Point> points, double learningRate)
{
double b_gradient = 0;
double m_gradient = 0;
double N = double.Parse(points.Count.ToString());
foreach (Point p in points)
{
//In our function:
/*
* y'=logy, b=log a, x'=log x
* When we found the values for b so the a will be retrieved: 10 power b = a
* Then we can make our formula: y=a x power m
*
*/
double pY = double.Parse(p.Y.ToString());
double pX = double.Parse(p.X.ToString());
//Remember: y'=logy, x'=log x
pX = Math.Log10(pX);
pY = Math.Log10(pY);
b_gradient += -(2 / N) * (pY - ((m_current * pX) + b_current));
m_gradient += -(2 / N) * pX * (pY - ((m_current * pX) + b_current));
}
TuningParamsClass tp = new TuningParamsClass();
tp.b = b_current - (learningRate * b_gradient);
tp.m = m_current - (learningRate * m_gradient);
return(tp);
}
private TuningParamsClass gradient_descent_runner(List <Point> points, double starting_b, double starting_m, double learning_rate, int num_iterations)
{
TuningParamsClass tpFinal = new TuningParamsClass();
double b = starting_b;
double m = starting_m;
for (int i = 0; i <= num_iterations; i++)
{
litOutput.Text += "<BR/><BR/>Step:" + i.ToString() + ", b: [" + b.ToString() + "] m: [" + m.ToString() + "] " + DateTime.Now.TimeOfDay.ToString();
Debug.WriteLine("Step:" + i.ToString() + ", b: [" + b.ToString() + "] m: [" + m.ToString() + "] " + DateTime.Now.TimeOfDay.ToString());
tpFinal = step_gradient(b, m, points, learning_rate);
b = tpFinal.b;
m = tpFinal.m;
litOutput.Text += "<br/><br/> > " + string.Format("error = {0}", compute_error_for_line_given_points(b, m, points));
Debug.WriteLine(" > " + string.Format("error = {0}", compute_error_for_line_given_points(b, m, points)));
}
return(tpFinal);
}
