Author: Tom Kerrigan
Date: 14:34:45 04/25/00
Go up one level in this thread
On April 25, 2000 at 14:03:05, Robert Hyatt wrote: >On April 25, 2000 at 11:27:34, Tom Kerrigan wrote: > >>On April 25, 2000 at 09:12:12, Robert Hyatt wrote: >> >>>On April 25, 2000 at 00:50:50, Tom Kerrigan wrote: >>> >>>>On April 24, 2000 at 22:13:10, Bruce Moreland wrote: >>>> >>>>>On April 24, 2000 at 18:49:04, Tom Kerrigan wrote: >>>>> >>>>>>On April 24, 2000 at 15:56:37, Dann Corbit wrote: >>>>>> >>>>>>>On April 24, 2000 at 15:43:24, Tom Kerrigan wrote: >>>>>>>[snip] >>>>>>>>Here's my question. If pondering=off cripples Crafty so badly to the point that >>>>>>>>Bob Hyatt has to write dozens of posts about it, why doesn't he just do >>>>>>>>something to fix it? I mean, surely the time spent writing all those posts could >>>>>>>>have been put to better use. >>>>>>> >>>>>>>That makes a great deal of sense if Dr. Hyatt were writing crafty to make people >>>>>>>happy who want to play engine verses engine matches on a single machine. >>>>>>>However, he does not play it that way and it is contrary to his purposes. >>>>>>> >>>>>>>Do you alter your programs to make them do what others wish even when it does >>>>>>>not coincide with your desires? >>>>>> >>>>>>Yes. >>>>>> >>>>>>-Tom >>>>> >>>>>If your point is that Bob should do this or that, I think that Bob should be the >>>>>one who decides what Bob does. It's great to suggest improvements in >>>>>functionality or support, but if Bob wants to do it his own way, that's fine. >>>> >>>>Decisions are influenced by your surroundings. >>>> >>>>Right now, Bob is surrounded by people who do matches between Crafty and ___ >>>>with no pondering. Consequently, Bob has to do a tremendous amount of damage >>>>control. Here are the options, as I see them: >>>> >>>>1) Continue to waste time by doing massive damage control >>>> >>>>2) Simply remove the ponder switch from Crafty, so Crafty can't be crippled >>>> >>>>3) Un-cripple Crafty >>>> >>>>Personally, I would not like to _continually_ make excuses for my program, i.e., >>>>option 1. I think option 2 is a hack, but still better than option 1. >>>>Personally, I would go with option 3. >>>> >>>>I don't really see what the problem is with option 3. If Crafty is using too >>>>much time in the opening and middlegame, just make it use less time. Multiply >>>>some number by 75% or whatever. It may not be a "fine tuned" solution, but at >>>>least the program won't lose all its games. >>>> >>>>-Tom >>> >>> >>>Once you add some sophistication to your time control logic, you will see that >>>the above is a very 'superficial' suggestion. Base time allocation is but one >>>part of the problem. How much time can you use (extra time) when you get a >>>positional fail-low, not a material one? How much extra time can you use on a >>>fail low for a single pawn? For a piece? What if you do a 12 ply search, and >>>the first 11 plies show you winning a pawn. At depth=12, after the first move, >>>you discover that move doesn't win that pawn. How much extra time do you use >>>there to see if the pawn win was real, or just a deep tactical plan by your >>>opponent that made the pawn a "phantom"... >> >>Why can't you just multiply all this stuff by 0.75? >> >>Anyway, I don't understand how pondering directly affects any of this. What is a >>situation where the 0.75 thing fails because pondering=off? And what's the >>"correct" thing to do in said situation? >> >>-Tom > > >Maybe it isn't a linear relationship between all the terms??? You don't have to have linear relationships to multiply by 0.75. Even if you used a fixed amount of time per move when ponder=off, wouldn't that be better than the current situation? I'm just saying, there has to be something quick and dirty that you can do so that Crafty isn't horribly crippled when people run ponder=off tests. -Tom > >I discovered, by _lots_ of testing (and help by Mike Byrne) that the current >'overflow' values work well with the normal engine setup. They burn a good >bit of time up front when it is needed, so that as correct predictions save time >later I don't end up in an endgame with a huge time surplus when it isn't >needed. But when pondering is turned off... _all_ of these values become >wrong. And in some games they produce severe time trouble. In others they have >hardly any noticable effect. > >But _any_ effect is too much effect...
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.