Author: Vasik Rajlich
Date: 15:05:13 03/15/04
Go up one level in this thread
On March 15, 2004 at 17:23:41, martin fierz wrote: >On March 15, 2004 at 15:28:55, Tord Romstad wrote: > >>On March 15, 2004 at 14:30:58, Vasik Rajlich wrote: >> >>>Yeah, actually the book was very good - but not for the reason Watson was hoping >>>for. You roll your eyes once per page when the "it's all about calculation" >>>comment inevitably comes up - and look at the games, conveniently organized by >>>the basic patterns ... :-) >> >>Is it a book you would recommend even for us lowly patzers, or is it the kind >>of book you won't understand a word of unless you are a really strong player? >> >>Tord > >i don't know how lowly a patzer you are, but it is certainly a thought-provoking >book. on the other hand, as an engine author you are probably looking for rules >to implement in your eval function, and you won't find any in this book - that >is watson's dogma - there are no rules... > >cheers > martin I partly agree. You won't get much that would help Gothmog, because what's useful for a human isn't useful for a computer. Consider the following "pattern": [d]r2q1rk/ppnn1pbp/3p2p/2pP/P3PP/2N2Q1P/BP4P/R1B2RK w - - Once you've seen this type of position as a human, you know the right plan: 16. e5 dxe5 17. f5 and black is suffering. A human learns this position by seeing a few instances of it, and his level gets higher as a result. However, this "knowledge" will be completely useless to a chess programmer. If you start trying to match this pattern inside your evaluation function, you'll just create a bloated mess, and it won't even do what you want. Instead, you program a few basic king safety terms, let your engine search a few ply, and voila! Vas
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.