He guys,

I have an idea for a software, and im pretty sure that somebody did something like this already.

In short: Clustering together loads of pcs to create a deep kibitzer.

A bit longer: You kibitz a match (the idea came to me when watching tcec, together with a 100 other ppl, all wondering ...

wow... thanks!

that saved me a headache

thanks to all of you. since im currently developing using gcc and the assembler there is just a bitch im postponing any such optimizations for now. when im returning to that point i know where to start now

some of you guys prbly did this already, but im wondering:
the LZCNT cmd in the SSE4a extensions (sadly only available on AMDs) offers a count of leading zeros. interesting for bitboards? so two questions:
- how would one force the c++ compiler of choice to use the LZCNT-command? code sample? im ...

some random thoughts concerning your experiment:
- i think it could be quite smth if done correctly. the parameters are, by definition, always pretty good. and if the trainingset is correctly chosen the resulting optimization could by mindblowing
- your lack of convergence over the two testsets ...

can u tell us a bit about your network and thoughts behind it? i am actually very much interested in using neural networks of various kinds for evaluation, but my background in this area is a bit slim. what did you do, which test-set, which settings, why backpropagation, what layout, etc?