Author: Robert Hyatt
Date: 15:06:41 01/04/99
Go up one level in this thread
On January 04, 1999 at 11:14:17, Leon Stancliff wrote: >On January 04, 1999 at 02:46:25, blass uri wrote: > >> >>On January 04, 1999 at 01:05:56, Peter McKenzie wrote: >> >>>On January 04, 1999 at 00:23:26, blass uri wrote: >>> >>>> >>>>On January 03, 1999 at 23:55:45, Robert Hyatt wrote: >>>> >>>>>On January 03, 1999 at 21:18:11, Leon Stancliff wrote: >>>>> >>>>>>Bob Hyatt, >>>>>> >>>>>>I think you could probably answer this question if anyone can. Do you know how >>>>>>the rating increases with ply depth. I have seen a figure of 200 points increase >>>>>>per ply. I do not believe this would be a constant figure. My Hiarcs 6 will >>>>>>normally reach 7 ply and sometimes 8 ply in the middle game. Do you have any >>>>>>kind of graph showing what might be expected with each ply increase from 4, 5, >>>>>>6, 7, 8 or 9 plies in the middle game. Does the increase in rating decrease as >>>>>>one moves to higher levels? In other words, do you get as more increase as you >>>>>>move from 7 to 8 than you do from 8 to 9. >>>>>> >>>>>>If anyone else has information on this topic please feel free to add your >>>>>>comments. >>>>> >>>>>A very hard question. I believe that "dumb" programs get less from another >>>>>ply than "smart" programs. At shallow depths, another ply helps _every_ >>>>>program avoid/find tactics. At deeper depths, tactics fade out somewhat and >>>>>if the program has little positional knowledge, it won't be able to get any- >>>>>thing out of the extra plies. A smart program will continue to improve, >>>>>however, as it simply finds deeper "plans" to use its positional knowledge. >>> >>>I've had similar thoughts, which have influenced the design of my own chess >>>program. This sort of thing is of course resonably hard to prove or disprove. >>> >>>>> >>>>>How you categorize the various programs would be fun however, as that would >>>>>start a small war here. My best guess would be to simply take a common >>>>>cpu and run all the programs. In general, the faster the NPS, the "dumber" >>>>>the program, since the eval is the best place to pick up speed (IE in Crafty, >>>>>I spend over 50% of my time in Evaluate() which is fairly high by today's >>>>>standard. Although programs like Hiarcs are probably at 75% or so, while a >>>>>program like Fritz is likely at 10% or less.) >>>> >>>>I do not think that Hiarcs earn more from time >>>>I read that Hiarcs7 is good at blitz. >>>> >>>>If you were right then I would expect hiarcs7 to have better results at >>>>tournament time control(relative to blitz) but I did not see it. >>> >>>Interesting point, what results are you basing this statement on? Is there a >>>blitz rating list similar to SSDF? >> >>I do not know but I know that Enrique did blitz tests Hiarcs7-Junior5 >>and Hiarcs7 was better. >> >>I also remember that he found that Hiarcs7 is a good solver of tactical >>positions. >>> >>>> >>>>Hiarcs7 won Junior5 at blitz and lost 6:4 against Junior5 at 40/40 games of >>>>enrique. >>> >>>I doubt very much that these results (10 game matches) are statistically >>>significant! >> >>I agre that they are not statistically significant after 90 games(eight 10 game >>matches against the same opponents and this 10 game match) >>Junior has 50.5 out of 90 and Hiarcs has 53 out of 90. >>not significant result at 40/40 >> >>I do not remember the result in blitz but I remember that Hiarcs7 was clearly >>better. >> >> >>> >>>> >>>>Uri >>> >>>Peter > >Let me get back to my original question to Bob Hyatt and be a little more >specific. Bob, How many plies does Crafty, as operating on the ICC, reach in the >middle game? How much of an increase would you expect from one more ply depth? >If 200 points is a valid figure at the lower levels and you can give me an >expected increase at the level Crafty is now playing, I can get a crude idea of >what happens from ply to ply. At least this would be true with respect to >Crafty. > >So far it appears that no one has done an investigation such as you suggested in >your first reply. > >Thanks, >Leon I'm not sure about depth... but I'd guess that in blitz games, the middlegame depth is in the range of 9 plies, maybe 10 (of course in some wild positions it might only be 8 or so as well). As far as what does a ply do for crafty? hard to quantify, but clearly every ply is important. IE the p6 was about 2.5X the P5/133 I used and its rating took a big jump. The quad p6 gave another factor of 3 and another noticable jump, and now the quad xeon (which ramps out at about 2x the quad p6, roughly) is another noticable jump. So far, I see *no* indication that another ply won't produce a better search result... what we need is for someone to organize a crafty vs crafty tournament, no pondering, and try one with sd=4 against sd=5,6,7,8,9,10 (at least). Then do sd=5 vs 6,7,8,9,10, then sd=6 vs sd=7,8,9,10 and so forth. That would give a good graph of what another ply (or more) is worth for crafty...
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.