Re: [gnugo-devel] breakdown of GnuGo 3.2 in Goban

[bump]

> While being challenged by GnoGo and its clones on NNGS at handicap of 
> 6 stones, I discovered Goban 4.8 for Mac OS X (Jaguar), and played
> a couple of games with it on my 400MHz G4 with 320 MB RAM
> running OS X 10.2.2.
> 
> In one game, it misread a ladder, quite far into the ladder, while taking
> about 3 minutes per move (at level 10) on the above machine.

What version of GNU Go is this? You can find out by running gnugo --version.

How did you make this sgf file? The handicap stones are added twice and
the file crashes GNU Go. I don't think this is legal sgf. I cleaned it up
and added it as regression/games/vaughn.sgf.

Unfortunately this is a current bug. I checked that the current CVS
version plays this way. I added a test based on this game at move 44 as
strategy5:278.

Of course the tactical reading code is very stable and would
not misread a ladder. However the owl code thinks this is
an owl defense. I haven't seen this bug before.

At move 44 we get the following traces:

  J11: owl attack/defend for J14
  J11: owl attack/defend for H16
  J11: owl attack/defend for F18
  J11: owl attack/defend for H18
  J11: owl attack/defend for E18
J11:   23.00 (followup) - threatens to capture G16
  J11: 60.68 - change in territory
  J11: 1.08 - strategic bonus for E18
  J11: 1.45 - strategic bonus for H18
  J11: 9.75 - strategic effect on D17
  J11: 0.35 - strategic effect on M17
  J11: 8.26 - strategic effect on G16
  J11:   5.50 - H16 strategic value already counted - A.
  J11:   3.30 - J14 strategic value already counted - A.
  J11:   26.33 - total followup value, added 3.33 as territorial followup
  J11: 13.16 - added due to followup (26.33) and reverse followup
  values (0.00)
  J11: 0.41 - connects strings (connect value 4, shape factor 0.082)
Move generation values J11 to 95.15

Dan