Author: Serge Desmarais
Date: 18:00:17 08/28/98
Go up one level in this thread
On August 28, 1998 at 18:03:38, Robert Henry Durrett wrote: >On August 28, 1998 at 16:36:29, Serge Desmarais wrote: > >>On August 28, 1998 at 12:55:57, Robert Henry Durrett wrote: >> >>>Why not obtain, and create a library of, the opening books of all of the current >>>chess engines [maybe others not so current too - maybe] and test all engines >>>with all of the books? >>> >>>This may be too big a project, so may not be practical for that reason alone. >>> >>>The thought behind this suggestion is that such a testing program would point >>>out any dependences on books as well as to make comparisons of the "engines >>>without their books" more informative relative to the true inherent strengths >>>and weaknesses of the engines. It might also point out highly specific needs for >>>improvements in some of the engines. >> >> >> You are assuming that the opening books are a kind of "cover up" of their >>respective program's weaknesses, if I understand you well? >> >>Serge Desmarais > >(1) Disclaimer: Please recall my previous "admission" that I am not a >professional chess software programmer. What I was thinking may or may not make >sense to a person with intimate knowledge of the inner workings of such >software. If not, I hope one of those people will squelch this idea now. >(2) I do not regard modern chess engines to be "weak" at all. They are, it >seems to me, very strong indeed. >(3) The purpose I had in mind was to unmask the basic [but very strong] engine >from the opening books so that any inherent "weaknesses" could more easily be >identified. >(4) Currently, any weaknesses which would cause the engine to not perform >perfectly in the early opening are masked by the opening book. I think it is more to gain free time to the clock. In a tournament game (40 moves in 2 hours)in which the move time is 3 minutes per move, staying for 20 moves in the book with no time spent, meanwhile, you end up with only 20 moves to play in 2 hours, so an average of 6 minutes for every move, thus doubling the number of positions seen, increasing depth and all the advantage, so playing a stronger chess. Also, if for most game positions, the program usually has a small preference for a certain move, the same is true for the starting position. So, it would tend to ONLY play THAT move it prefers. That may be 1.e4 or 1.Nf3 or 1.d4 which ARE good moves (some HUMAN players do only play ONE move as White in the starting position!) thus only playing ONE line of play against most Black's replies, for example. Creativity is not a strong point of computers. Along with the loss of strenght, caused by having to "think" on the very first move, these are good reasons for opening books to exist. Of course, if your program is quite weak in CLOSED positions, you could tweak it to favor or exclusively play open positions. But you cannot only put lines that are good for the program because most users use the opening books to study the opening theory. So, you also have to put all the major lines for all the major openings, PLUS the major lines for most minor openings. And not all the programs are well optimized. Fritz, for example, is having a lot of problems while playing the French, still nothing in its book prevents it from playing it. Serge Desmarais >(5) But people here on CCC have recently posted some very good reasons why chess >engines should not be entered into competitions "naked" i.e. without an opening >book. >(6) The competitions between engines are both competitions in inherent strengths >of the basic engines [without books] and competitions in development of better >books. >(7) Keep in mind that programs like Fritz have rating increases of only a few >points every now and then. The threads dealing with statistics have shown that >it takes quite a few games to determine with any confidence that such a rating >increase is "real." >(8) When you are dealing with very small rating increases, as noted in another >thread, it is unclear whether this increase was due to an improved book or to an >improvement in the way the engine arrives at it's next move when out of book. >(9) So, maybe I am thinking that the books "cover up" engine performance >weaknesses, as you say.
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.