Author: Robert Hyatt
Date: 09:19:39 05/26/04
Go up one level in this thread
On May 26, 2004 at 02:10:19, Joel wrote: >It is an interesting phenomonen! > >Good fundamental algorithm/poor optimisation is one probably the main cause, but >here are some additional things I believe play a part (at least with my engine): > >- long time controls very much hide the bad time management by bodo > >- i believe my hashing replacement scheme is better than a lot of engines at my >strength level, so if we play longer time controls, then the other engine's hash >table can get saturated, causing a move ordering penalty. > >- my nps is not as high as a lot of other engines, so longer time controls >lessen the chance of being many plies of search behind... > >- i use ETC and IID, and at my level, a lot of engines don't have these features >which potentially (certainly do for me) make a difference at longer time >controls. > >- weak evaluation of my engine compared to others of similar strength. the >search saves me from many poor moves. > >Although this list is hardly conclusive, they are the main points which spring >to my mind at the moment. I find it significant that they all seem directly >related to a 'poor' job by the programmer. > >Joel There are other issues. A very simple q-search can be a very large part of a blitz search, but it is less an issue when you reach depths of 14+ plies in long games. Null-move R=3 can be a killer at shallow depths as it hides lots of tactics that result in major problems. At depths of 14+ it is much harder for this same sort of failure to occur and still be forced. IE there is a good probability that if My PV has moves A, B and C, and I play A and my opponent plays B, then I am reasonably forced to play C in many positions. But if the PV goes far deeper in a 14 ply search, moves 8-10 plies down the PV are unlikely to be played at all in the real game in most positions. So errors are hidden by opportunities to vary farther down the PV.
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.