Author: Christophe Theron
Date: 12:24:00 04/24/00
Go up one level in this thread
On April 24, 2000 at 05:14:53, Mogens Larsen wrote:
>Since noone is willing to comment on these questions. I'm going to make an
>attempt even though I have very little knowledge about chess programs.
>
>On April 23, 2000 at 15:50:22, Christophe Theron wrote:
>
>>Several points to think about:
>>
>>1) What is the only authorized time control to use to decide program's strength?
>>Who decided this?
>
>The most favoured time control is the one used in most tournaments. Thereby it's
>the most likely scenario for a computer program to encounter. So it's not who
>but what.
Most of the games played on earth are played at faster time controls.
>>2) What is the time control most used in reality by chess programs users?
>
>I try to use a timecontrol where the chess programs get at least 30 seconds per
>move. There are exceptions to this rule of course. The reason for my choice is
>the availability of my computer. If I had two it wouldn't be a problem to choose
>longer timecontrols.
I think generally users use time controls faster than tournament time controls.
>>3) Where is your data about the difference in playing strength between blitz
>>and, say, 40 moves in 2 hours, of any program, on a significant number of games?
>
>Non existent, but I would assume that anyone fond of chess likes to see the best
>move played within "reasonable time". At blitz the best moves are less frequent
>unless the selectivity is very good.
I wonder. I remember a test suite published in a french magazine some years ago.
The computers of that time had to think for several minutes in order to solve
the problems. And these problems came from grandmaster games (far over the level
of even a good club chess player).
Current computers can solve all the problems in only a few seconds. Most
problems are solved in less than 1 second.
>>4) When you play a blitz on a very very fast computer, it is like playing the
>>same game at slow time controls on a much slower computer. What does "blitz"
>>mean in a world where you can find very fast and very slow computers?
>
>On average there isn't that big a difference in speed. Not everyone owns a quad.
>As far as I understand the relationship between hardware and strength isn't
>linear, so is there a problem? I use longer timecontrols because my computer is
>very slow.
There are a lot of older PCs still in use. People don't buy a PC every year.
Around me I see people still using 486dx2-66 or Pentium 90 computers. These are
more than 9 times slower than the K6-2 450 computers that are currently used by
the SSDF. They are more than 16 times slower than Athlon 800 computers.
A tournament game on P90 (40 moves in 2 hours) means 40 moves in 13 minutes on a
K6-2 450. 40 moves in 8 minutes (approx) on Athlon 800.
Look at the Pentium 90 part of the SSDF list. The games played to compute this
list have been played at blitz rate for an Athlon 800.
Did someone say that they are meaningless?
>>5) What is the difference in search depth between a blitz game and a slow time
>>control game? Why do you expect that one program will benefit more than its
>>opponent from this deeper search?
>
>I assume there's a big difference in emphasis on search depth, selectivity and
>chess knowledge between the different programs. This should warrant a difference
>in time preference, but I'm no expert.
The difference in search depth between blitz and tournament time controls is 4
plies in average.
Top programs already manage to search 8-9 plies deep in blitz.
8-9 plies in blitz and 12-13 plies at tournament time controls. No big deal.
>>6) What evidence do you have that the curve of "relative strength vs time
>>control" is monoton? I mean it is possible that prog A is better than prog B at
>>blitz, then prog B is better at 40 moves in 2 hours, then prog A is again better
>>at move in 1 day. If you take for granted that prog A wins at blitz, but would
>>lose at 40 moves in 2 hours, you cannot reject my proposition so easily. And
>>which program is stronger in this case?
>
>If you use the reference to human games. Then program B would be the strongest
>player.
Why using the reference to human games?
For correspondance games you might prefer the other one.
>>The assumption that the relative strength of chess programs changes with the
>>time controls used is one of the many legends that people like to believe in.
>>
>>It might or might not be true for a given pair of programs, but it is NOT a
>>general rule.
>
>I don't think anyone claims that it is a general rule. Maybe it's just nice for
>the tester to think about the game simultaneously. It might be easier to check
>the evaluation process of your program as well.
>
>I tried to answer your questions, but I know nothing about computer chess
>programs except some of the things explained in this forum. I hope my effort
>hasn't been totally futile (there it was again :o)).
Thanks, I felt a little bit sorry that nobody wanted to discuss these issues.
Christophe
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.