Author: Robert Hyatt
Date: 09:45:59 10/30/99
Go up one level in this thread
On October 30, 1999 at 09:58:57, Enrique Irazoqui wrote: >On October 30, 1999 at 09:49:14, Robert Hyatt wrote: > >>On October 30, 1999 at 09:32:22, Peter Schneider wrote: >> >>>Time-management of programs dependent of remaining >>>time of the opponent? >>> >>>In human-human matches the time for reflection is considerably >>>influenced from the remaining time of the opponent. >>>E.g., if the opponent has only little time at his clock (Zeitnot), >>>fast moves can be beneficial in order to shorten his >>>"permanent brain" time. >>> >>>Is this concept already realized in programs? >>> >>>Peter Schneider >> >> >>Mike Byrne wrote code like this for Crafty. We ran with it for a year or >>so and it had its plusses. But on the other side of the coin, you are ahead >>on time, and yet you use less because your opponent has less. I finally >>decided that I didn't like the overall result, although the idea seems >>reasonable. > >I think it was Tal who in his Autobiography said that playing quickly when the >opponent was in time trouble was a typical mistake of amateur players. Instead, >he recommended to use time fully, but complicating matters as much as possible. > >Enrique Correct. The problem I saw was that this worked well against humans (more on this later). But against computers, it is easier to make a mistake they will spot if you don't use more time than they do. Several years ago I ran an interesting experiment at an ACM event. I think this was around 1992, but I am not certain. We always played some blitz games vs gm players, if we had any machine time left over after the tournament game was over. This particular year, we basically had access to the machine all night every night. The common way to play humans at 'blitz' was that the human would get 5 mins on the clock, the computer would get 5 minutes of total think time, and a limit of 60 moves max. This was because the computer operator had to type in moves, read the output and make moves, etc. Everyone played like this for many many years. Including us. Playing like that, we were beating Ivanov about 3 of every 4 games, and had similar results against other GM/IM players. Which wasn't bad. However, I decided to try something new, because I was a reasonable chess player and had no problem handling move input/output 'mental conversions' and after being a systems programmer for 30 years, I can type very quickly as well. So I decided to play using the "real" clock for Cray Blitz also, no more 5 secs/move for 60 moves. I was going to trust my typing. The time control I used was this: CB used 3 secs / move until it had used 1 minute of time total, then it dropped to 2 secs/move until it had used 2 minutes of time total, then it dropped to 1 sec/move for the remainder of the game. The first thing that happened was that humans couldn't just 'hang on' until move 60 and claim the win on time as was happening before. The second thing is that it appeared that the program was moving nearly instantly when you factor in pondering. This tended to get the human player into a faster rhythm. And after I made that change, we didn't lose another single game for the remainder of the event. Ivanov came back for one final marathon and I believe he drew 1 game out of 20, with no wins. When I asked him, Valvo, Kopec, and others, they all felt that it put more 'pressure' on them by moving quickly, and they were very concerned by its tactical accuracy at that speed. So against humans, this might not be bad. But at 40/2hr games, I don't think I would even consider trying it...
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.