[gnugo-devel] Valuation problem

[bump]

I've been working on Evan's patch evan_5_1.1a on the double
kakari and I encountered the following difficulty.

(;GM[1]FF[4]
SZ[19]
GN[GNU Go 3.4 load and print]
DT[2003-08-21]
KM[0.5]AP[GNU Go:3.4]RU[Japanese]
;AW[nc][oc][pc][qd][oe][pe][qe][qf][pg]
AB[qc][dd][nd][od][pd][ne][nf][of][pf][qg][qh][dp][pp]
PL[W])

This is not joseki. In the joseki line W cuts at O14 before
connecting at P17, but I've been testing moves close to but
not identical to the joseki to see how robust GNU Go's
play is in such a position. The answer is, not very.

GNU Go wants to play Q12:

Actually in many related positions GNU wants to play Q12
but this one is egregious because the stones at R16 are
owl critical.

  Q12: owl attack/defend for Q13
  Q12: owl attack/defend for R16
  Q12: 51.38 - change in territory
  Q12: 5.92 - strategic effect on O16
  Q12: 7.56 - strategic effect on R13
  Q12:   11.46 - total followup value, added 11.46 as territorial followup
  Q12: 5.73 - added due to followup (11.46) and reverse followup values (0.00)
  Q12: -8.68 - shape (shape values +0.00(0) -5.00(2), shape factor 0.746)
  Q12: 0.63 - connects strings (connect value 6, shape factor 0.126)
Move generation values Q12 to 62.53

Needless to say, Q12 is not a good owl defense for R16. For
one thing, it doesn't work. For another, even if it did
work Q13 is light and should not be defended until R16 is
settled. The only move in this position is S17.

Dan