Looking for volunteers/help for a reasonbly easy Encog extension... PROBEN1

Okay, if you've been looking for a chance to contribute to Encog, this might be it. This falls very much under the domain of traditional programming, and not so much under the more "advanced mathematical" extensions we've been doing with Encog lately. Also it is fine to help in C# or Java, this is very much a "parallel" effort. Also I will guide/help this effort, and write the "core" of it.

Here is the idea. Encog has been benchmarked fairly extensively. But this is mostly in terms of how fast can Encog calculate a backpropagation iteration compared to some of the other frameworks out there. We have very little in the way of benchmarking how well a training iteration does. This is really more useful for benchmarking Encog against itself! Why do such a thing? Maybe you want to know what network structure or activation function works best. For me, I am doing quite a bit with extending Encog NEAT, and there are a whole "witches brew" of extensions to NEAT to consider, and who knows which ones actually help vs harm. It would be nice to have a very standard Encog benchmark that measures both time AND error accuracy.

Encog does benchmark accuracy in some areas. Like weight initialization. But almost always we use a Fahlman Encoder. Which is a good enough way I suppose, but an encoder is FAR from representative of real data.

There is actually a standard benchmark. It is called PROBEN1. I would like to add a full PROBEN1 evaluation to Encog. That works with ANY ML method. I.e. we could pit NEAT vs ANN, or ANN vs SVM or anything. There are 15 tests in PROBEN1. I will take a few and create the core of this "example", then others can extend this and add the remaining. I will assign the tests, and answer any questions. This is actually a very good introduction to how to get data into and out of Encog.

Would any one be interested in helping? Advanced knowledge of AI/Math is defiantly NOT needed for this one.