Author: Robert Hyatt
Date: 07:47:40 10/14/03
Go up one level in this thread
On October 14, 2003 at 10:13:06, Peter Fendrich wrote:
>On October 14, 2003 at 09:47:38, Robert Hyatt wrote:
>
>>On October 13, 2003 at 14:08:51, Christophe Theron wrote:
>>
>>>On October 13, 2003 at 08:31:04, Anthony Cozzie wrote:
>>>
>>>>On October 13, 2003 at 08:16:20, Djordje Vidanovic wrote:
>>>>
>>>>>Hi,
>>>>>
>>>>>while preparing the opening book for Ruffian I decided to use a very good
>>>>>positional program for Ruffe's sparring partner. I decided on Diep due to its
>>>>>impressive positional play. Diep also has an interesting and unorthodox opening
>>>>>book with lots of lines that are worth analysing. No small wonder, the book's
>>>>>creator is a super strong Fide Master, the author of Diep: Vincent Diepeveen.
>>>>>
>>>>>Be it as it may, I matched Ruffian with only a skeleton of the book to be
>>>>>(meagre 1538 positions for starters) and pitted the positional monster against
>>>>>the fast searcher. The result was a little disappointing and I must say that I
>>>>>did not learn much from the match. Of course, bear in mind that these were only
>>>>>G/5 games, but still...
>>>>>
>>>>>Diep had its own rather well researched book, with many home cooked tricks and
>>>>>traps, while Ruffian was equipped with a wee book that is to grow yet. Diep had
>>>>>the advantage of a Barton 2800+ while Ruffian played on my old NetVista PIII-933
>>>>>computer.
>>>>>
>>>>>End result: Ruffian 86%, Diep 14%, or 48-8!! My question is: could the
>>>>>reigning leader of the SSDF beat Diep more convincingly than Ruffian?
>>>>
>>>>Two things come to mind:
>>>>
>>>>1. I didn't look at all the games, but it looks like Diep opened every game 1.
>>>>Nh3??
>>>>
>>>>2. Diep is more designed for longer time controls. I remember Vincent
>>>>complaining last CCT about how 60 10 was too short ;)
>>>
>>>
>>>
>>>TMUEAGAB (The Most Used Excuse After Getting A Beating, tm)
>>>
>>>I do not know if the setup of this match is correct and if Diep is really so
>>>weak, but I know that asking for longer time controls is just a way to spread
>>>fog.
>>>
>>>If a chess program really needs longer time controls to start playing decently,
>>>then there is something inherently wrong in its design.
>>>
>>>In other words, it sucks.
>>>
>>>I'm not saying that Diep sucks. Maybe the match setup was not fair for it.
>>>
>>>I'm saying that if a program gets such a beating at blitz it does not smell good
>>>anyway for a longer time controls match.
>>
>>
>>Some examples:
>>
>>program A evaluates forks. Program B does not. Which will do better at
>>very shallow/fast searches? Program A will avoid forks while program B will
>>walk into them. As the time stretches out, program A's advantage will
>>shrink. I saw this happen in the 1970's.
>>
>>Program A does a parallel search. That _clearly_ gets better as depth
>>increases (not without bound, but the difference between a parallel 3 ply
>>search and a parallel 10 ply search is huge.) Program A might get killed at
>>very shallow searches because of that performance problem.
>>
>>King safety is another issue. A program that does this better will perform
>>better at fast time controls. At longer time controls a good search can
>>compensate somewhat for king safety.
>>
>>I've previously explained a similar issue that almost cost me the 1986 WCCC
>>title, because I tuned on a slow machine but ran on a way faster machine. The
>>normal version lost way more games to a micro on very slow hardware, but the
>>"new and improved version" won nearly all from the micro. But put the new and
>>improved on the Cray and it played so incredibly passively that it simply did
>>poorly, period.
>>
>>It isn't hard to tune for a specific depth and use eval to fill in holes that
>>the tactical search can't handle. That program will out-play the same program
>>without that specific eval term, until the depth is great enough that the search
>>compensates.
>>
>
>Ahh!
>So you are defending Diep and Vincents view now! :-)
>
>/Peter
I'm only saying that it is possible to have a program that favors shallow
depths over deep depths. Not that it is reasonable to do so. I happen to
agree with Christophe in that a _good_ program should probably do well at
both fast and slow time controls.
I'm simply pointing out that it is certainly possible that a program
does worse at fast time controls. Or better.
That's hardly the same as agreeing with Vincent when he _always_ complains
that the time control was too fast for him. In my case, I believe that my
program does better at longer time controls. But only because that is where
I do all my tuning.
One example. How do you tune/limit extensions? It is possible to tune them
so that they are most efficient at one specific depth. Go deeper and they
tend to over-extend, go shallower and they tend to under-extend. Yes, you
should be able to tune it so it does the correct thing for all cases, but
I don't necessarily believe that I have done that. I am sure I _could_ do
that, given the time, however.
>
>>
>>
>>>
>>>That's incredible. I hear the same excuse ("it will perform better at longer
>>>time controls") since the days of the 386. Now that our computers are several
>>>hundred times faster, the same excuse is still used. It does not make any sense.
>>>
>>>
>>>
>>> Christophe
This page took 0.01 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.