Author: Dann Corbit
Date: 12:08:42 03/08/06
Go up one level in this thread
On March 07, 2006 at 22:54:09, Vikrant Malvankar wrote: >On March 07, 2006 at 22:20:35, Dann Corbit wrote: > >>On March 07, 2006 at 22:12:42, Nathan Thom wrote: >> >>>On March 07, 2006 at 21:34:10, Dann Corbit wrote: >>>> >>>>Opening books hold the frequently played positions near the origin. >>>>If we are talking about some position 50-60 plies down the road, the odds of >>>>hitting it during game play are "astronomical". No, they're "commical" -- and >>>>uneconomical. >>>> >>>>Just the bare positions -- ignoring half-move clock and 3-time repeat are 10 to >>>>the 50th power. So, let's suppose that our intrepid programmer analyzes one >>>>billion positions. The odds in hitting one of them are one in ten to the >>>>forty-first power. Not good. Plus you would have a bit of bloat storing the >>>>positions and a bit of time spent searching for them. >>>> >>>>Now, let's suppose that we bypass all these objections and say "What the heck, >>>>let's do it anyway!" >>>> >>>>Well, when we look at memory, we will see (one billion * hash element size) >>>>bytes of memory consumed. A very small hash entry would consume 16 bytes but >>>>we'll say he's clever and stores only 8 bytes. That would be 8 gigs of ram. >>>> >>>>"Well..." (you may retort) "perhaps they are loaded on demand." >>>> >>>>I suppose that a page fault for every new position would slow down the program >>>>so much that we would see 50-100 NPS at best. While Rybka may be a slow >>>>searcher (let's not start that debate) it's certainly not that slow. >>>> >>>>I suppose we're just going to have to admit that V.R. is a clever guy, and that >>>>he hasn't stored the middle game in the computer's data segments. >>> >>>What about only considering parts of the board (<64 squares) or only specific >>>pieces. e.g. only consider rooks+kings and have a pre-generated table of the >>>most common situations and best move? Sure, the other pieces which have been >>>ignored could make the move ridiculous or illegal but i wonder what kind of >>>success rate this would give? >> >>In the evaluation it would give good success. But that is what everyone does. >>As data statements it would have zero usefulness. > >Are current Engines using differant evaluation techniques based on the move no. >and pieces remaining on the board? Yes, with different levels of participation in that idea. Even the most basic programs have "game phase" as part of the eval.
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.