As i understand, the optimization parameters are selected for robustness (by selecting the best average PF in the "neighborhood", not just at a single parameter value), from the Training a Strategy page (http://zorro-trader.com/manual/) - had a couple of questions -

1) If there are 2 parameters, does using a 2-dimensional map of Profit Factors to determine neighborhood of PFs improve robustness significantly, or is the two 1 dimensional maps (each of which contain the average PF for each dimension) a good approximation of robust neighborhoods?

2) What's the metric/formula used for selecting the robustness?
Is it something like
Robust Par1 = Max(Average(PF(i-1), PF(i), PF(i+1)) where i is the parameter step, assuming 10% steps => you're looking at a neighborhood of +/-10% of parameter range?

thanks
Kiran

1) Depends on the system. Robust systems normally don't have several peaks and valleys in the parameter space, so two 1-dimensional maps are the best solution. However if the parameter space is very irregular with many peaks, 1-dimensional maps won't converge to the highest peak.

2) As to my information, spline interpolation is used for finding the most robust point on the spline.

Hi,

what kind of spline is used? There are many splines possible: BSpline, BezierCurve, LoessInterpolation, CubicBSpline and so on

Also some of them have settings associated with them, which ones do you use?

Would it be possible to draw the spline into the parameter result chart/histogram? Or is the source for the spline available somewhere so I can just plot it manually?

Best regards,
subes