[gnugo-devel] attack either valuation bug

[Gunnar Farneback]

Here's a bug which needs fixing.

Look at the NNGS game gnugo-3.3.2-Programmer-200206021522.sgf, move
52. Clearly J14 is the only move, and at level 10 GNU Go finds this.
However, the move was played at level 4, which resulted in an
unreasonable tenuki at Q17. It's interesting to see the valuation of
J14 then:

  J14: 0.0 - (threat of) attack/defense of F16 (owl attack/defense as well)
  J14: 0.0 - (threat of) attack/defense of H15 (owl attack/defense as well)
  J14: 0.0 - (threat of) attack/defense of K18 (owl attack/defense as well)
  J14: owl attack/defend for K18
  J14: owl attack/defend for H15
  J14: owl attack/defend for F16
  J14: owl attack/defend for K19
  J14: 49.37 - change in territory
  J14: -8.58 - either attacks H17 (-8.58) or attacks L17 (-1.83)
  J14: -8.58 - either attacks M18 (-3.00) or attacks H17 (-8.58)
  J14: -8.58 - either attacks J18 (-8.58) or attacks L17 (-1.83)
  J14: -8.58 - either attacks J18 (-8.58) or attacks M18 (-3.00)
  J14: -8.92 - either attacks K19 (-8.92) or attacks L17 (-1.83)
  J14: -8.92 - either attacks K19 (-8.92) or attacks H17 (-8.58)
  J14: -8.92 - either attacks K19 (-8.92) or attacks M18 (-3.00)
  J14: -8.92 - either attacks K19 (-8.92) or attacks J18 (-8.58)
  J14:   2.45 - K15 strategic attack/defend
  J14:   7.65 - H15 cut/connect strategic value
  J14:   2.85 - M18 cut/connect strategic value
  J14:   10.27 - K19 cut/connect strategic value
  J14:   5.93 - F16 cut/connect strategic value
  J14:   8.10 - H15 cut/connect strategic value
  J14:   3.20 - L17 cut/connect strategic value
  J14:   3.26 - K15 cut/connect strategic value
  J14:   13.75 - K18 cut/connect strategic value
  J14: 10.00 - K18 cutstone
  J14:   4.28 - M18 attacked/defended
  J14:   4.80 - L17 attacked/defended
  J14:   3.26 - K15 attacked/defended
  J14:   10.27 - K19 strategic value already counted - B.
  J14:   13.75 - K18 strategic value already counted - A.
  J14: 4.28 - strategic effect on M18
  J14: 4.80 - strategic effect on L17
  J14:   5.93 - F16 strategic value already counted - A.
  J14:   8.10 - H15 strategic value already counted - A.
  J14: 3.26 - strategic effect on K15
  J14:   3.75 - total followup value, added 3.75 as territorial followup
  J14: 1.87 - added due to followup (3.75) and reverse followup values (0.00)
  J14: 0.37 - connects strings (connect value 5, shape factor 0.104)
Move generation values J14 to 3.95

Obviously there's something very fishy about the negative
contributions from the attack_either move reasons. Presumably this
could be fixed just by making sure that adjusted_worm_attack_value()
never returns negative values, but there might be more to this. Does
someone feel like investigating what's really going on here?

/Gunnar