[gnugo-devel] Strategic effect

[bump]

Inge wrote:

> But that is not the biggest problem. The biggest problem is that the
> strategic effect gets assigned in proportion to dragon2.weakness.  Of
> course, the score should be proportional to the *change* in weakness
> if the move is played.  Is there a way to calculate this cheaply?

Probably not. But we need to see what we can do about assigning
strategic effect more intelligently. In some cases it needs to
be increased, and in some cases it needs to be decreased. There
are a lot of clues in the information we do collect, and we need
to figure out how to read them.

As an experiment I tried doubling the strategic effect for owl
attack/defend move reasons, awarded at line 2136 of value_moves.c. 
As Inge would have predicted this gave more unexpected passes
than fails. But my hope is that by examining these results we
can get some insight into how to tune the strategic effect.

I think it might make sense to increase the strategic effect
for owl defense reasons but not for owl attack reasons. I will
try this next. The reason is that I often see GNU lose groups
by not defending or expanding a weak group and this seems to
me a more serious problem right now than lack of aggressiveness.

It would be good to be able to differentiate between light stones
and heavy stones. Strategic effect should not be awarded for light
stones like K10 in trevorb:250. The stones at L8 in nngs1:8 are
heavy and defense is urgent.

Dan

./regress.sh . trevora.tst 
510 unexpected PASS!
./regress.sh . nngs1.tst 
4 unexpected FAIL: Correct 'H3|J5', got 'S14'
5 unexpected FAIL: Correct 'H3|J5', got 'S14'
7 unexpected PASS!
8 unexpected FAIL: Correct 'M10|L9', got 'S14'
26 unexpected FAIL: Correct 'P3', got 'O6'
30 unexpected PASS!
31 unexpected PASS!
35 unexpected PASS!
46 unexpected PASS!
48 unexpected FAIL: Correct 'R16', got 'F15'
52 unexpected PASS!
./regress.sh . strategy.tst 
20 unexpected PASS!
./regress.sh . arion.tst 
2 unexpected FAIL: Correct 'Q18|S16', got 'E7'
./regress.sh . ego.tst 
8 unexpected PASS!
./regress.sh . trevorb.tst 
250 unexpected FAIL: Correct 'C2', got 'L8'
340 unexpected FAIL: Correct 'M11', got 'F3'
./regress.sh . strategy2.tst 
70 unexpected FAIL: Correct 'M15|J15|N16|J16', got 'C15'
83 unexpected PASS!
89 unexpected PASS!
94 unexpected FAIL: Correct 'A2', got 'J2'
./regress.sh . nicklas1.tst 
502 unexpected FAIL: Correct 'G5', got 'H7'
./regress.sh . niki.tst 
4 unexpected PASS!
./regress.sh . nngs.tst 
690 unexpected FAIL: Correct 'N13', got 'O10'
1800 unexpected FAIL: Correct '!H5|G4|F6|F7', got 'G4'
./regress.sh . trevorc.tst 
520 unexpected FAIL: Correct '!C5', got 'C5'
600 unexpected PASS!
./regress.sh . global.tst 
5 unexpected FAIL: Correct 'O4', got 'F5'
31 unexpected FAIL: Correct 'M8', got 'D11'
./regress.sh . arend.tst 
35 unexpected FAIL: Correct 'H17|J19', got 'F15'
38 unexpected FAIL: Correct 'O5', got 'O17'
./regress.sh . 13x13.tst 
28 unexpected FAIL: Correct 'L10', got 'E3'
./regress.sh . trevord.tst 
490 unexpected FAIL: Correct 'D5', got 'G4'
520 unexpected FAIL: Correct 'D15', got 'G16'
860 unexpected FAIL: Correct 'Q5', got 'N4'
870 unexpected PASS!
./regress.sh . strategy4.tst 
194 unexpected FAIL: Correct 'A13', got 'D6'
200 unexpected FAIL: Correct 'P6|P7|Q7|S18', got 'S9'
216 unexpected PASS!
./regress.sh . handtalk.tst 
3 unexpected PASS!
./regress.sh . strategy5.tst 
225 unexpected PASS!
./regress.sh . century2002.tst 
220 unexpected PASS!
270 unexpected PASS!
./regress.sh . safety.tst 
2 unexpected PASS!

nngs1:4,5. Acceptable. Defense of R13 is certain urgent though O13 seems
           better than the hane at S14.
nngs1:8.   Bad. At this point S14 not as important as M10.
           This is a good position to tune: both O13 and M10 should
           be preferred to S14.

For nngs:8 we get:

     L9: 2.43 - change in territory
     L9:   19.17 - L8 uncertain owl defense bonus
     L9: 19.17 - strategic effect on L8
     L9:   2.97 - total followup value, added 2.97 as territorial followup
     L9: 1.48 - added due to followup (2.97) and reverse followup values (0.00)
     L9: 0.41 - connects strings (connect value 4, shape factor 0.082)
   Move generation values L9 to 23.50
     O13: 6.29 - change in territory
     O13:   11.70 - Q13 strategic attack/defend (eff size 11.50, weakness 0.68
     O13: 11.70 - strategic effect on Q13
     O13:   3.14 - total followup value, added 3.14 as territorial followup
     O13: 1.57 - added due to followup (3.14) and reverse followup (0.00)
     O13: 2.00 - shape (shape values +2.00(1) -0.00(0), shape factor 1.102)
   Move generation values O13 to 21.55
     S14: 6.41 - change in territory
     S14:   11.70 - Q13 strategic attack/defend (eff size 11.50, weakness 0.68
     S14:   0.83 - R14 strategic attack/defend (eff size 4.25, weakness 0.13
     S14: 11.70 - strategic effect on Q13
     S14: 0.83 - strategic effect on R14
     S14:   0.00 - total followup value, added 0.00 as territorial followup
     S14: 7.71 - shape (shape values +7.00(1) -0.00(0), shape factor 1.407)
     S14: 0.20 - connects strings (connect value 2, shape factor 0.040)
   Move generation values S14 to 26.84

   Top moves:
   1. S14 26.84
   2. L9  23.51
   3. O13 21.56
   4. Q11 20.67
   5. M10 20.45
   6. R11 20.14
   7. K16 17.26
   8. C13 17.06
   9. C7  17.04
   10. F17 16.74

---------------------------------------------------------------------

nngs1:26.  Why are O3 and P8 amalgamated? Strategic effect of O6
           on P8 and Q10 is real. Strategic effect on O3 is not.
     O6: 7.68 - change in territory
     O6:   2.02 - P12 strategic attack/defend (eff size 29.00, weakness 0.05)
     O6:   10.56 - O3 strategic attack/defend (eff size 25.50, weakness 0.28)
     O6: 2.02 - strategic effect on P12
     O6: 10.56 - strategic effect on O3
     O6:   2.74 - total followup value, added 2.74 as territorial followup
     O6: 1.37 - added due to followup (2.74) and reverse followup (0.00)
     O6: 0.20 - connects strings (connect value 2, shape factor 0.040)
   Move generation values O6 to 21.83
   Top moves:
   1. O6  21.84
   2. C2  20.53
   3. P3  19.00
   4. Q3  18.53
   5. R2  17.64
   6. Q2  17.49
   7. G17 17.07
   8. E13 15.14
   9. L18 14.33
   10. E12 14.24


---------------------------------------------------------------------

nngs1:48. F15 is extremely large but not 48 points. Joseki is G15 but
          in this area F15 is also huge. It gets strategic value for three
          dragons amounting to a huge valuation.

     F15: 11.76 - change in territory
     F15:   12.74 - D16 strategic attack/defend (eff size 11.83, weakness 0.72
     F15:   8.20 - D14 strategic attack/defend (eff size 16.17, weakness 0.34
     F15:   3.85 - F17 strategic attack/defend (eff size 8.75, weakness 0.29
     F15: 12.74 - strategic effect on D16
     F15: 3.85 - strategic effect on F17
     F15: 8.20 - strategic effect on D14
     F15:   9.78 - total followup value, added 9.78 as territorial followup
     F15: 4.89 - added due to followup (9.78) and reverse followup (0.00)
     F15: 6.53 - shape (shape values +3.00(1) -0.00(0), shape factor 1.158)
     F15: 0.20 - connects strings (connect value 2, shape factor 0.040)
   Move generation values F15 to 48.18
   Top moves:
   1. F15 48.18
   2. R16 37.22
   3. F16 28.38
   4. E17 24.84
   5. Q14 24.81
   6. R8  23.61
   7. O6  23.36
   8. G15 22.55
   9. J3  20.86
   10. C8  17.53

---------------------------------------------------------------------

arion:2. E7 is slightly overvalued but locally it makes more sense
         to atari at E8, a move not considered.

     E7:   7.17 (followup) - threatens to capture D8
     E7:   18.10 - B4 strategic attack/defend (eff size 19.08, weakness 0.63)
     E7: 18.10 - strategic effect on B4
     E7:   7.17 - total followup value, added 0.00 as territorial followup
     E7: 3.58 - added due to followup (7.17) and reverse followup (0.00)
     E7: 1.08 - shape (shape values +1.00(1) -0.00(0), shape factor 1.050)
   Move generation values E7 to 22.77
   Top moves:
   1. E7  22.77
   2. S16 22.44
   3. B16 20.29
   4. H4  18.64
   5. J3  18.55
   6. J4  17.20
   7. F6  15.14
   8. K4  14.25
   9. O14 13.74
   10. P12 11.51

---------------------------------------------------------------------

trevorb:250. Neither the strategic effect on K4 nor the strategic
             effect on K10 is very appropriate. Both are light stones.
             Defense is not urgent---not in the same way that
             the defense of the two weak groups in nngs1:8. Meanwhile
             C2 is urgent.
trevorb:340. This is a mirror image of trevorb:250. We don't need both.

For trevorb:250 we get:

     L8: 12.33 - change in territory
     L8:   12.49 - K4 strategic attack/defend (eff size 15.25, weakness 0.55)
     L8:   10.24 - K10 strategic attack/defend (eff size 17.50, weakness 0.39)
     L8: 12.49 - strategic effect on K4
     L8: 10.24 - strategic effect on K10
     L8:   9.46 - total followup value, added 9.46 as territorial followup
     L8: 4.73 - added due to followup (9.46) and reverse followup (0.00)
   Move generation values L8 to 39.79
   Top moves:
   1. L8  39.79
   2. C2  37.10
   3. C4  34.23
   4. L7  24.64
   5. K8  16.93
   6. H11 16.72
   7. K7  16.60
   8. K6  15.90
   9. F11 15.88
   10. G10 15.42

---------------------------------------------------------------------

strategy2:70. C15 is overvalued due to two strategic effects,
              of which the effect on D17 is unwarranted. C15
              threatens to capture and cut off something but
              defense of the top is urgent.
   Top moves:
   1. C15 35.82
   2. N16 34.59
   3. E13 27.28
   4. P16 25.27
   5. M15 19.93
   6. M19 19.51
   7. G14 14.99
   8. F16 14.57
   9. J16 14.43
   10. J15 14.31
     C15: 1.50 - change in territory
     C15: 7.00 - strategic effect on D17
     C15: 5.13 - strategic effect on C17
     C15:   14.13 - total followup value, added 14.13 as territorial followup
     C15: 7.07 - added due to followup (14.13) and reverse followup (0.00)
     C15: 14.71 - shape (shape values +10.00(2) -0.00(0), shape factor 1.710)
     C15: 0.41 - connects strings (connect value 4, shape factor 0.082)
   Move generation values C15 to 35.82