[gnugo-devel] Semeai in 3.1.17

[Daniel Bump]

                       Semeai in 3.1.17

In 3.1.17 several tests are broke in the experimental semeai
version. Some of these are real semeais, most of them are
not. A problem is that I don't really have a good criterion
for when a semeai is real or not. These examples will show
what I mean. I'd be glad to get opinions about how to
solve this problem.

Certainly if both dragons have more than one hostile neighbor
the semeai should probably be considered real. However this
sufficient criterion is not necessary as the upper right
corner in the three positions from bretz.sgf shows.

I'll refer to certain situations as faux semeai. Often
these are situations in which one or the other dragon is
morally inessential. A possible resolution of such a situation
would be that both dragons have status DEAD, and the 
inessential dragon ends up as captured stones inside a
dead dragon. For example in strategy 37, even if the
(morally inessential) stones at B18 do get cut off and
die, the B dragon cannot live. Oddly enough the stones
at B18 are not owl insubstantial because placing B
stones around them to remove them from the board
results in two eyes, the second one at the 1,1 point.

Accurately characterizing inessential stones is an
important problem. However it may be easier to
characterize genuine semeais directly than those
which are not genuine.

                      REAL SEMEAIS.
13x13 25
gnugo -l gnugo-3.1.18/regression/games/mertin13x13/katsunari-gnugo3.B+14.sgf -L 
64
   This example involves real semeais. A semeai is found between J3 and K5. 
   The matcher status of J3 is changed from CRITICAL to DEAD and the matcher 
   status of K5 is changed from CRITICAL to ALIVE. The move at L4 is chosen.
   This is a little strange if J3 is thought dead. There is another large
   semeai on the left side of the board which seems to get ignored. (A7
   is not small.)
http://www.public32.com/regress/?13x13:48

nicklas1 1405:
   A genuine semeai found between D9 and B8. Changed the matcher status of 
   D9 from CRITICAL to ALIVE. This is a misreading.
gnugo -l gnugo-3.1.18/regression/games/nicklas/nicklas14.sgf -L 55 -d0x040
http://www.public32.com/regress/?nicklas1:1405

strategy4 168
   This is a real semeai problem. A semeai is found between D8
   and A6. The matcher status of B4 is changed from CRITICAL
   to DEAD. The matcher status of A6 is changed from DEAD to
   ALIVE. This is erroneous since B4 can live independent of
   the semeai at A4.
gnugo -l gnugo-3.1.18/regression/games/strategy40.sgf -L 30 -d0x040
http://www.public32.com/regress/?strategy4:168

strategy4 197
   A semeai is found between N18 and Q18. No statuses are
   changed, but T18 is recommended. This is consistent with
   Trevor's comment "tenuki at S18 is not unthinkable."
   S18 is actually wrong since it probably allows seki.
gnugo -l gnugo-3.1.18/regression/games/bretz.sgf -L 130 -d0x040
http://www.public32.com/regress/?strategy4:197

strategy4 199
   A semeai is found between N18 and Q18. No statuses 
   are changed but the move at T18 is found. Another
   semeai is found between C1 and E3. No statuses are
   changed. The move at T18 is selected
gnugo -l gnugo-3.1.18/regression/games/bretz.sgf -L 164 -d0x040
http://www.public32.com/regress/?strategy4:199

strategy4 200
   The semeai between N18 and Q18 is again found. Another
   semeai is found between R14 and R8. The status of
   R14 is changed from CRITICAL to ERROR (presumably
   UNKNOWN). The move at Q8 is selected. The semeai
   between R14 and R8 is certainly bogus. This would
   seem to show that the two semeai dragons must be
   in physical contact.
gnugo -l gnugo-3.1.18/regression/games/bretz.sgf -L 212 -d0x040
http://www.public32.com/regress/?strategy4:200

                    FAUX SEMEAIS

strategy 37: 
  A (faux) semeai is found between B18 and D19. The owl
  code has already correctly read out that D19 cannot be made
  to live, but the semeai module changes its status from
  DEAD to CRITICAL.
gnugo -l gnugo-3.1.18/regression/games/nicklas/nicklas8.sgf -L84 -d0x040
http://www.public32.com/regress/?strategy:37

arb 104: 
  A (faux) semeai is found between B8 and B9. The matcher
  status of B8 is changed from CRITICAL to ALIVE and the matcher
  status of B9 from CRITICAL to DEAD. This is nonsensical since
  B8 is tactically critical. This position should be handled
  by the atari_atari module. 
gnugo -l gnugo-3.1.18/regression/games/arb/game02.sgf -L71 -d0x040
http://www.public32.com/regress/?arb:104

golife 4:
  Faux semeais are found between J6 and J5 and between H4 and J5.
  Changes the matcher status of H4 from CRITICAL to ALIVE,
  and the matcher_status of J5 from CRITICAL to DEAD.
gnugo -l gnugo-3.1.18/regression/games/golife.sgf -L32 -d0x040
http://www.public32.com/regress/?golife:4

nicklas1 501:
   A faux semeai between E7 and F7. Changes matcher status of E7 from CRITICAL
   to ALIVE, and the matcher_status of F7 from CRITICAL to DEAD.
gnugo -l gnugo-3.1.18/regression/games/nicklas/nicklas5.sgf -L23 -d0x040
http://www.public32.com/regress/?nicklas1:501

nicklas2 1401
   A faux semeai is found between B1 and B2. The matcher status of B1 is
   changed from DEAD to CRITICAL. The move at A2 is found.   
   Trevor comments that fighting ko is not unreasonable so A2 may be
   considered a correct answer, too.
gnugo -l gnugo-3.1.18/regression/games/nicklas/nicklas14.sgf -L 31 -d0x040
http://www.public32.com/regress/?nicklas2:1401

strategy3 126
   A faux semeai is found between C19 and E19. Of course C19 
   should be regarded as inessential. The matcher status
   of C19 is changed from CRITICAL to ALIVE. The matcher status
   of E19 is changed from CRITICAL to DEAD.
gnugo -l gnugo-3.1.18/regression/games/strategy29.sgf -L 138 -d0x040
http://www.public32.com/regress/?strategy3:126

global 16
   A faux semeai is found between P13 and P11. The matcher status of P11
   is changed from CRITICAL to ALIVE. (It seems to me that this should
   not be regarded as a true semeai but that is perhaps open to debate.)
gnugo -l gnugo-3.1.18/regression/golois/Goemate990903-6.sgf -d0x040
http://www.public32.com/regress/?global:16


                           OOPS!

13x13 48
   This test does not appear to involve the semeai module. It
   only appears in this list because the test result was not
   correctly updated during the 3.1.17 release.
gnugo -l gnugo-3.1.18/regression/games/mertin13x13/gnugo-goliath2.W+38.sgf -L 
11 -d0x040
http://www.public32.com/regress/?13x13:48