BGonline.org Forums

Upcoming release speed improvements

Posted By: eXtreme Gammon
Date: Sunday, 29 May 2011, at 3:05 p.m.

In Response To: Upcoming release speed improvements (Michael Depreli)

Q: Can you tell us the pruning method used?
Sorry, this is "trade secret" :)

Q: Will the user be able to switch off pruning?
No, i dont think i will give that option as the difference is so little. Allowing it would require that i change all the display (screen, export etc) to display if prune was used or not.

Q: How many games were tested pruning v non pruning and what were the largest errors with pruning on? Do you have any examples?
I made many tests. I reran all the different level throught the 1000 test games. The differences of level range from +0.2 to -0.5, with an average across the level of -0.18 Elo (0.006 PR)

I also test a set of 20,000 positions. the avergage difference between prune and no prune is 0.0007+/-0.0020. Comparing to RO data:

No prune: error=0.0113+/-0.0110

Prune: error=0.0114+/-0.0111

The repartition of difference is

less than 0.001: 82.74%

between 0.001 and 0.002: 8.21%

between 0.002 and 0.003: 3.68%

between 0.003 and 0.004: 1.86%

between 0.004 and 0.005: 1.04%

between 0.005 and 0.010: 1.86%

between 0.010 and 0.020: 0.50%

more than 0.020: 0.13%

Note that 45% of time the pruned evaluation is actually closer to RO results than the no Prune one.

here is the position with the most difference:

Prune is better:
• XGID=-----BC-B----BA-A--B--fcB-:0:0:-1:00:0:0:2:0:0 NP=1.728 P=1.799 RO=1.814
• XGID=---A--BBB----AAA-----hAfD-:0:0:-1:00:0:0:2:0:0 NP=1.122 P=1.178 RO=1.239

Prune is worse:
• XGID=-BCB-BC-A-aBb------ccbca--:0:0:-1:00:0:0:2:0:0 NP=0.593 P=0.629 RO=0.588
• XGID=-a-BCCCBB---a----a-bbbbbb-:0:0:-1:00:0:0:2:0:0 NP=0.850 P=0.878 RO=0.846
• XGID=--BBCBBa------A--bcbcC-bb-:0:0:-1:00:0:0:2:0:0 NP=-0.522 P=-0.495 RO=-0.551

Messages In This Thread

• Can't get XG add a move function to work
  Michael Nielsen -- Friday, 27 May 2011, at 12:48 a.m.
  • Can't get XG add a move function to work
    eXtreme Gammon -- Friday, 27 May 2011, at 2:13 a.m.
    • Can't get XG add a move function to work
      Michael Depreli -- Friday, 27 May 2011, at 6:16 a.m.
    • Can't get XG add a move function to work
      Michael Depreli -- Saturday, 28 May 2011, at 10:44 a.m.
      • Can't get XG add a move function to work
        Michael Petch -- Saturday, 28 May 2011, at 1:49 p.m.
        • Can't get XG add a move function to work
          Michael Petch -- Saturday, 28 May 2011, at 1:59 p.m.
          • Can't get XG add a move function to work
            Michael Nielsen -- Saturday, 28 May 2011, at 2:07 p.m.
            • Fixed for the next release
              eXtreme Gammon -- Saturday, 28 May 2011, at 3:24 p.m.
              • Fixed for the next release
                Michael Depreli -- Saturday, 28 May 2011, at 5:45 p.m.
                • Upcoming release speed improvements
                  eXtreme Gammon -- Saturday, 28 May 2011, at 6:57 p.m.
                  • Upcoming release speed improvements - WOW!!! O_O *NM*
                    mtuhtan -- Saturday, 28 May 2011, at 7:57 p.m.
                    • Upcoming release speed improvements - WOW!!! O_O
                      neilkaz -- Sunday, 29 May 2011, at 5:19 a.m.
                  • Upcoming release speed improvements and computers
                    neilkaz -- Sunday, 29 May 2011, at 5:30 a.m.
                  • Upcoming release speed improvements
                    Michael Depreli -- Sunday, 29 May 2011, at 7:42 a.m.
                    • Upcoming release speed improvements
                      eXtreme Gammon -- Sunday, 29 May 2011, at 3:05 p.m.