Log in

PNNeric using Parzen window classification

Hi I found here one of the most interesting EAs written by the Russians implementing for a Probabilistic Neural Network implementing a Parzen Window Classification. As far as I understand it is using also a kernel method.

This expert was published on TSD but did not find much interest. The code is extremely interesting.

The Parzen windows classification algorithm does not require any training phase

What matters here to be optimizsed is the parameter sigma. The width $\sigma$ of the Guassian determines the relative weighting of near and far points. Tuning this parameter controls the predictive power of the system. We have empirically optimized the value of $\sigma$.

But it is a pain because  the lack of sparseness makes the test phase quite slow.

And yes it is extremenly slow to be optimized in the mt4 optimizer. But it is basically plug and play.

 

Comments