Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: DB will never play with REBEL, they simple are afraid no to do well

Author: Ratko V Tomic

Date: 23:50:14 10/16/99

Go up one level in this thread


>> 1. d4 d5 2. c4 e6 3. cxd5 exd5 4. Bf4 Bb4+ 5. Nc3 Nf6 6. e3 Nc6 7. Bb5
>
> 9->  16.35
>10->  46.00  (EBF from 9->10 = 2.81)
>11->   2:46  (EBF from 10->11 = 3.61)
>12->   9:39  (EBF from 11->12 = 3.49)
>Average EBF is 3.30333...
>
>Jeremiah

Now I see that Fritz UI shows total nodes (for all iterations to a given depth).
So I subtracted total nodes (and total times) shown for successive depths to get
nodes evaluated at given depth setting, and here are the figures I get with
Crafty 16.6 (on Celeron 400Mhz, 64Mb hash):

 Depth Tot.kN Delta.kN   BF    Time[s]  kN/s
 --------------------------------------------
   8    2243    ....    ....     11     ...
   9    4705    2462    5.13     23     205
  10   11557    6852    4.83     57     202
  11   38582   27025    4.74    189     205
  12  119442   80860    4.56    579     207
 --------------------------------------------

So the branching factors are 4.6-5.1, a bit larger spread than the earlier one
(because now I included 8->9 which is the largest here, the others didn't change
much). So it seems we get quite different BF. What version is your Crafty and
does it show total nodes evaluated? Could you provide the node figures? I have
used all default settings.

Now that I was at it, I also tested Fritz 5.32, Hiarcs 7.32 and Rebel 10c on the
same position. For Hiarcs & Rebel I didn't do depth 12, since 11-ply already
took too long. Rebel and Hiarcs selected Bd3 move while Fritz and Crafty
selected Bb5 (which was my move in the game), at all levels for all programs.

Fritz 5.32:

 Depth Tot.kN Delta.kN   BF    Time[s]  kN/s
 --------------------------------------------
   8    1198    ....    ....      3     ...
   9    5117    3919    5.40     14     356
  10    6651    1534    4.16     18     383
  11   19361   12710    4.42     54     353
  12  227861  208500    4.93    640     355
 --------------------------------------------

 Hiarcs 7.32:

 Depth Tot.kN Delta.kN   BF    Time[s]  kN/s
 --------------------------------------------
   8     801    ....    ....     21     ...
   9    2475    1674    4.91     65      38
  10    9831    7356    4.86    259      38
  11   46176   36345    4.87   1218      30
 --------------------------------------------

 Rebel 10c:

 Depth Tot.kN Delta.kN   BF    Time[s]  kN/s
 --------------------------------------------
   8    3243    ....    ....     22     ...
   9    6353    3110    5.26     45     135
  10   18652   12209    5.11    135     136
  11   56967   38405    4.89    410     140
 --------------------------------------------


So none is getting BF close to 3. Fritz shows the largest jumps, it seems it has
special optimization/shortcuts for particular lower depths. But Crafty seems to
catch up and work faster at depth 12. Hiarcs has the most uniform branching
factor, suggesting that it was set manually in the form of some forward pruning
parameter, e.g. maybe it has width=5, plus some shortcuts acting more randomly
which make it slightly below 5. Its work per node is about 10 times larger than
for Fritz. Rebel has the largest BF, pretty close to exact minimaxing, but which
seems to decrease uniformly with depth, which is suggestive of some forward
pruning depending on depth (i.e. larger pruning at larger depths). In terms of
work per node, Rebel is between Fritz and Hiarcs, which is also true for its
style of play.



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.