Error definition in Encoq

Diamantregen's picture

Hi,

A quick question, how is the ITrain.Error defined? Somehow it doesn't seem to be the MSE (mean square error). Thanks.

Regards,
Roger

jeffheaton's picture

Basically, I take the square of every error difference. Then at the end I take the square root of that divided by the number of cases.

Diamantregen's picture

Hi Jeff,

Thanks for the info. Uhm, is there any chance to add a method for MSE error (http://www.informatik.htw-dresden.de/~iwe/Belege/2004/Matzke/Dokumentati... and http://www.informatik.htw-dresden.de/~iwe/Belege/2004/Matzke/Dokumentati...)? I could imagine that it could be done without much costs, especially as it's very similar to the root mean square. Last but not least, MSE is the most commonly used error calculation method for ANNs. Thanks.

Regards,
Roger

Diamantregen's picture

Hi,

I have digged a little into the Encog code and really think it's feasible to add MSE support. However in the meantime I will just use a quick&dirty approach which is modifying the ErrorCalculation class this way and rebuilding Encog:

[...]
public double CalculateRMS()
{
double err = this.globalError / this.setSize;
return err;
}
[...]

Regards,
Roger

Diamantregen's picture

.


Copyright 2005 - 2012 by Heaton Research, Inc.. Heaton Research™ and Encog™ are trademarks of Heaton Research. Click here for copyright, license and trademark information.