Re: Search-Based Opening Book Construction
Posted: Sat May 06, 2017 4:59 am
You could collect all those positions(there shouldn't be too many of them) and subject them to deep analysis. You might find lines thought to be bad that are good, or you could find flaws in your evaluation.jdart wrote:I have been thinking about taking my existing book, evaluating all the terminal positions and using minimax values in move selection. Then possibly using dropout expansion. Currently it correctly avoids a lot of bad lines and selects good ones, but the move weights are probably not very optimal. I have not done this yet. There would be some decisions to make - for example should moves manually flagged as "don't play" be kept that way or should you actually let the eval do the selection or should you use some weighting of prior information and eval?
--Jon