Author: Robert Hyatt
Date: 17:39:40 05/18/03
Go up one level in this thread
On May 18, 2003 at 04:28:08, scott farrell wrote: >On May 16, 2003 at 10:42:25, Robert Hyatt wrote: > >>On May 15, 2003 at 21:47:20, Jon Dart wrote: >> >>>Ernst Heinz did this by using standard test suites, for example Win at Chess, or >>>ECM (Encyclopedia of Chess Middlegames). He found that the solve rate didn't >>>really change much with forward pruning on, but the number of nodes searched for >>>a fixed ply depth decreased 20-50% (this is from the chapter on AEL pruning in >>>his book Scaleable Search in Computer Chess). He also used other testing >>>methods, including game play, as detailed in the book. >> >> >>I don't think this methodology is reasonable. >> >>For example, suppose your forward-pruning speeds you up by a factor of >>four. Comparing same-search-depth runs means the FP version will move >>in 1/4th the time of the non-FP version. Suppose the FP version does >>worse on three positions? But if you run it for 4x longer, so that it >>has the same time limit as the non-FP version, the three positions are now >>"back to normal"??? >> >>I think the _right_ way to test is with a fixed time limit so that a version >>that can go deeper will go deeper. After all, that is the _purpose_ of FP >>in the first place... > >maybe you can run the non-fp version, and time how long it takes to find the >solution. Rhen allow a maximum time control for the fp version equal to the >solve time - if it solves in less, success, if it doesnt solve in time = >failure. > >Scott That's my point. I use a "sum of squares" approach. I search to solution, square the time and sum all such times for all the positions. I compare this to the sum of squared times for the other version, and the version with the "smaller" sum of squares is better... > >> >> >> >> >>> >>>--Jon >>> >>>On May 15, 2003 at 18:17:09, Russell Reagan wrote: >>> >>>>I would like to know how to test whether or not a forward pruning method is >>>>reliable. >>>> >>>>I have one idea to test when and if a method is reliable, and I'd like to know >>>>if it's a good idea or not, and also what other methods might be used to test >>>>the reliability of forward pruning methods. >>>> >>>>My idea requires a collection of games, and two versions of a program. One >>>>version would have forward pruning turned off, and the other would have it >>>>turned on. You would feed each version of the program the same game, and let >>>>each do a search on the initial position to the same fixed depth. If both >>>>versions report the same move and score, and the version using forward pruning >>>>had a lower time to depth, then the forward pruning is reliable (so far). If the >>>>version using forward pruning reported different results, then the forward >>>>pruning method is not reliable for this type of position. You make the next move >>>>in the game, and repeat the search and compare the results for each position in >>>>the game. Then you repeat the process for each game. >>>> >>>>When I think about testing the reliability of null-move using this method, I >>>>think the test would do well. I would expect the test to tell us that in most >>>>positions, null-move is reliable, and I would expect it to fail for some endgame >>>>positions, and so this test would tell us that null-move was good forward >>>>pruning, but to turn it off in the endgame (or detect zugzwang, or however you >>>>choose to guard against it). I haven't had time to test this though, since I >>>>just thought of it and I'm not at home. >>>> >>>>I am basing all of this on the assumption that the strength forward pruning >>>>provides is not that it finds better moves at the same depth, but that it >>>>finishes searching a particular depth in a shorter amount of time, allowing the >>>>search to go deeper, which is where the added strength comes from. Is this >>>>correct? >>>> >>>>Comments, please...
This page took 0.01 seconds to execute
Last modified: Thu, 07 Jul 11 08:48:38 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.