Re: [gnugo-devel] games against 3.3.16

[SP LEE]

> 
> Recently it was reported here that 3.3.19 did poorly against
> 3.3.16, so I collected 95 games between the current CVS and
> 3.3.16. It seems that or 3.3.16 does a little better. 
> 
> In these games 3.3.16 is Black and the komi is 5.5. Black
> won 52 of the games, or 54% of the games, and the mean
> score is 2.12 points in favor of Black.
> 
> I'm attaching the 66 point game at the end. I didn't see anything
> too unexpected in White's play, just the kind of same kind
> of mistakes that we've been seeing in every game. W happened
> to lose a group early and never recovered.
> 
> So although this twogtp run is a little disappointing I don't
> think it's a showstopper for 3.4.
> 
> Game records available on request.
> 
> Dan
> 

Here is little more additional information on the move 48 of the game. The OWL 
analysis is good, telling O18 is the defense point. However, the tactical 
reading gives N18 and seems to be overvalued.

  N18: defense of worm M18
  N18: -2.03 - suspected ineffective defense of worm M18
  N18:   5.33 (followup) - threatens to capture L18
  N18: 15.36 - change in territory
  N18:   6.40 - total followup value, added 1.07 as territorial followup
  N18: 3.20 - added due to followup (6.40) and reverse followup values (0.00)
  N18: 0.63 - connects strings (connect value 6, shape factor 0.126)
  N18:   35.00 - minimum accepted value
Move generation values N18 to 35.00

Top moves:
1. N18 35.00
2. O18 28.31
3. M19 14.89
4. K11 14.55
5. J10 14.14
6. J9  12.73
7. H10 12.35
8. J11 12.35
9. E13 10.70
10. J6  10.68
Checking safety of a white move at N18
Move generation likes N18 with value 35.00
genmove() recommends N18 with value 35.00
= N18

I have a strong feeling that we should trust OWL analysis more in many cases.

SP Lee