Author: Mikael Bäckman
Date: 00:08:53 03/18/04
Go up one level in this thread
On March 17, 2004 at 18:19:49, Dann Corbit wrote: >On March 17, 2004 at 16:28:30, Mikael Bäckman wrote: > >>On March 17, 2004 at 16:22:42, Dann Corbit wrote: >> >>>On March 17, 2004 at 16:14:40, Mikael Bäckman wrote: >>>> >>>>Hi, >>>> >>>>I ran some tests with history and different methods of aging it, as discussed in >>>>this thread: >>>> >>>>http://www.talkchess.com/forums/1/message.html?354812 >>>> >>>>I picked 15 test positions from this years Linares tournament. 10 positions are >>>>from move 20 and 5 from move 40. I didn't study the positions much before >>>>selecting them. >>>> >>>>I used 90 seconds per position as I didn't know how deep I could search without >>>>spending days on this... First I ran a test without historytables, to get a >>>>depth to compare the other tests to. Most of the depths were completed in 20-60 >>>>seconds. Perhaps a bit shallow, but it gives an idea of the performance. >>>> >>>>I use a side-piece-to historytable or history[side][piece][to] and I use at most >>>>8 history moves at a node. After that I try the moves in the order they are >>>>generated. >>>> >>>> >>>>Test1 = No History >>>>Test2 = History >>>>Test3 = History - root aging >>>>Test4 = History - age as soon as a history score gets larger than 10000. >>>>Test5 = Same as 2 but with pawnmoves generated after all other moves. Included >>>>this for fun, but it seems to work best. :) >>>> >>>>Aging was done by dividing the values in the history tables with 8. >>>>Nodecounts are in thousands. >>>> >>>>Pos D Test1 Test2 Test3 Test4 Test5 >>>>-------------------------------------------------- >>>>01 12 27960 22044 21481 21923 21954 >>>>02 12 37488 31165 31706 25631 26492 >>>>03 10 34388 24471 24652 29455 24225 >>>>04 12 25099 21307 23497 20555 23460 >>>>05 13 30665 22026 22288 22156 21798 >>>>06 10 16141 12861 13447 13050 13576 >>>>07 14 44136 32362 31157 32776 32958 >>>>08 14 39848 38378 39681 38337 28706 >>>>09 11 31083 21410 24811 25470 25403 >>>>10 12 38152 29568 28020 29394 25669 >>>>11 13 29184 25017 27149 24854 23437 >>>>12 13 52650 27674 24784 26427 25901 >>>>13 14 58192 38986 41854 37978 41428 >>>>14 13 50823 45372 41400 41283 45473 >>>>15 13 63876 33226 32296 33651 32625 >>>>-------------------------------------------------- >>>> 579685 425867 428223 422940 413105 >>>> (136%) (100%) (101%) (99%) (97%) >>>> >>> >>>A 40% improvement for history heuristic is well above average improvement, I >>>think. >>> >>>Did you have hashing and other move ordering techniques in play? I think not. >>>If you have hashing operational, I would expect less than 25% improvement. >> >> >>Yes. >>I sort moves by: >> >>1. Hash >>2. Good captures >>3. Equal captures >>4. Killers >>5. History >>6. Losing captures >> >>I use SEE for captures. >> >> >>>What happens when you have all of your move ordering techqniques in use and then >>>you change only the history heuristic? >> >>All moveordering techniques were in use. The above happens. :) > >That is very surprising. Do you use all hashed pv nodes where depth is >= >target depth? Yes. >A 40% reduction is very striking. A possible explanation is that I don't use IID, nor do I store hash after a successful nullmove, and also I use a single depth preferred hashtable. I suspect my hashhit % is a bit lower than most engines, which makes history more important for me. /Mikael
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.