Author: Anthony Cozzie
Date: 06:23:25 08/02/04
Go up one level in this thread
On August 01, 2004 at 12:54:03, Steffen Basting wrote: >Hi Gerd! > >Thanks for replying! > >>Mate in 4 should be independent from search depth. >>Isn't the game over after Rh8#? > >Yes, of course :-) > >>Seems that you don't terminate your PVs correctly. > >I've just fixed this. It now says: >0.00s 1= 315 187 f4f1 >0.00s 2= 1926 148 f4f1 d7c7 >0.04s 3= 23053 131 f4f2 e6g4 e1e8 a8e8 d4a7 >0.20s 4= 93395 89 e1d1 f7f6 g5h4 a8c8 >0.87s 5= 515342 97 f4f1 f7f6 g5f4 e6f5 e1c1 >2.48s 6= 1452346 68 f4f1 e6f5 c2c3 e8e1 f1e1 a8e8 e1d1 e8e2 >6.68s 7= 4129525 70 f4f2 e6g4 e1f1 f7f6 g5f4 d7e6 f4c7 >8.57s 8= 5185655 120 f4f2 e6g4 e1f1 f7f6 g5f4 g6g5 f4d6 d7d6 >8.58s 8+ 5195575 370 f4f2 e6g4 e1f1 f7f6 g5f4 g6g5 f4d6 d7d6 >8.60s 8+ 5205409 870 f4f2 e6g4 e1f1 f7f6 g5f4 g6g5 f4d6 d7d6 >8.73s 8++ 5296335 #4 d4g7 g8g7 g5f6 g7g8 f4h4 e6f5 h4h8 # >9.09s 9= 5529524 #4 d4g7 g8g7 g5f6 g7g8 f4h4 e6f5 h4h8 # >9.64s 10= 5896253 #4 d4g7 g8g7 g5f6 g7g8 f4h4 e6f5 h4h8 # >11.86s 11= 7560467 #4 d4g7 g8g7 g5f6 g7g8 f4h4 e6f5 h4h8 # >19.04s 12= 12961258 #4 d4g7 g8g7 g5f6 g7g8 f4h4 e6f5 h4h8 # >34.34s 13= 23696937 #4 d4g7 g8g7 g5f6 g7g8 f4h4 e6f5 h4h8 # > >>How to translate your mate scores? > >When a mate occurs, I return 20000+depth so that "short" mates are preferred. >But you were right, I forgot to set the counter to 0 then, this caused those >strange lines. > >>1.bugs > >Yeah, there are probably plenty of them :-) > >>2.better eval > >Does eval have such a positive effect on depth / node ratio? > >>3.carefull with R==3 near the leaves, checks in qsearch? > >Yes, R=3 seems to be dangerous sometimes. I now use R=2 when the depth is >low. I don't use checks in qsearch (yet)... > >>4.SEE or more sophisticated move ordering, checks, putting pieces enprise, >> at interior nodes one may use more expensive move scoring/sorting. >>5.hashing score and bounds or flags as well. >>6.fractional extension/reductions. >>7.tuning. > >4-7 will be done after I've finished with 1 ;-) > >Thanks again, >Steffen I think eval has two effects on move ordering. First, a good eval decreases your flip-rate, which helps with parallel search and reduces the amount of time the engine "doubts", which is expensive. However, because it increases the granularity it also reduces the number of "=" beta cutoffs. Unfortunately I think the second effect tends to dominate. anthony
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.