Re: [gnugo-devel] Patch: Strategic effect for strategic attacks and defenses

[bump]

> I made another small experiment, using Dan's new code in the simplest
> possible way. The result is pretty mixed, and analyzing it might give
> useful hints for improving both approaches to measuring escape potential.

I too did some testing last week with some version of 3.3.9. I did
this last week while the FSF was flooded out by heavy rains in
Boston and the list was down for several days, so I'm only reporting
it now.

I tried tweaking the strategic bonus by taking into account the surround
status of a dragon. The testsuite consists of the following tests.

10 trevora:200
20 nngs1:35
30 nngs1:47
40 nngs1:48
50 nngs1:50
60 nngs1:52
70 strategy:50
80 trevorb:390
90 trevorb:450
100 trevorb:830
110 strategy2:51
120 strategy2:64
130 strategy2:70
140 strategy2:83
150 strategy2:90
160 strategy2:91
170 manyfaces:1
180 niki:4
190 trevor:220
200 buzco:3
210 nngs:1670
220 nngs:2020
230 trevorc:600
240 trevorc:1080
250 trevorc:1550
260 trevorc:1650
270 strategy3:101
280 global:25
290 global:33
300 arend:35
310 13x13:28
320 13x13:149
330 13x13:65
340 trevord:230
350 trevord:870
360 strategy4:200
370 strategy4:216
380 handtalk:19
390 nngs2:400
400 nngs2:480
410 nngs3:290
420 nngs3:800
430 century2002:120
440 century2002:440
450 safety:1
460 safety:4
470 safety:7
1010 trevora:220
1020 strategy:20
1030 neurogo:12
1040 trevorb:790
1050 strategy2:70
1060 strategy2:89
1070 strategy2:94
1080 nngs:690
1090 arend:38
1100 13x13:41
1110 trevord:120
1120 trevord:490
1130 strategy4:201
1140 nngs2:240
1150 nngs2:290
1160 nngs3:140
1170 nngs3:320

I have also available specific information about each case
in the following form. For test 20, i.e. nngs1:35 the
strategic effect is invoked as follows:

strategic defense D13 of nonsurrounded C12
strategic defense D13 of nonsurrounded D16
strategic defense D17 of nonsurrounded D16
strategic attack D17 does not surround B16
strategic defense F4 of nonsurrounded D4
strategic defense F4 of nonsurrounded H3
strategic defense H17 of nonsurrounded Q17
strategic defense N15 of nonsurrounded Q17
strategic attack N15 does not surround Q16
strategic defense P9 of nonsurrounded P10
strategic attack P9 does not surround O8
strategic attack P9 does not surround P12
strategic defense P16 of nonsurrounded Q17

I won't post that information but I have it if somebody wants it.

I considered weakly surrounded to be not surrounded.

The results are reported in the following format:

= A B C D E F

The value 1.8 at line 1943 of value_moves.c is replaced by:

A - strategic defend move, surrounded dragon, move inside surround field
B - strategic defend move, surrounded dragon, move outside surround field
C - strategic defend move, dragon not surrounded
D - strategic attack move, surrounded dragon
E - strategic attack move, nonsurrounded dragon, move surrounds
F - strategic attack move, nonsurrounded dragon, move does not surround

My intuition is that the strategic effect should be higher for B and E.
Yet for this testsuite, I found little sensitivity for those two parameters.

70 unexpected PASS!
1170 unexpected PASS!
= 2.00 1.80 1.80 1.80 1.80 1.80

60 unexpected FAIL: Correct 'L5', got 'L15'
320 unexpected FAIL: Correct 'L7', got 'L9'
330 unexpected PASS!
= 1.60 1.80 1.80 1.80 1.80 1.80

70 unexpected PASS!
1170 unexpected PASS!
= 2.20 1.80 1.80 1.80 1.80 1.80

= 1.80 2.00 1.80 1.80 1.80 1.80

= 1.80 2.20 1.80 1.80 1.80 1.80

= 1.80 1.40 1.80 1.80 1.80 1.80

450 unexpected FAIL: Correct 'R6|S2', got 'O14'
460 unexpected FAIL: Correct 'G14|E13|F13|C17', got 'L14'
= 1.80 1.20 1.80 1.80 1.80 1.80

1070 unexpected FAIL: Correct 'A2', got 'J2'
1100 unexpected PASS!
1110 unexpected FAIL: Correct 'B9', got 'F7'
1140 unexpected FAIL: Correct 'F17', got 'B14'
1150 unexpected PASS!
1170 unexpected PASS!
= 1.80 1.80 2.00 1.80 1.80 1.80

1140 unexpected FAIL: Correct 'F17', got 'B14'
1150 unexpected PASS!
= 1.80 1.80 1.90 1.80 1.80 1.80

50 unexpected FAIL: Correct 'O10', got 'N17'
70 unexpected PASS!
140 unexpected FAIL: Correct 'P14', got 'H17'
300 unexpected PASS!
= 1.80 1.80 1.60 1.80 1.80 1.80

40 unexpected PASS!
50 unexpected FAIL: Correct 'O10', got 'N17'
70 unexpected PASS!
140 unexpected FAIL: Correct 'P14', got 'H17'
180 unexpected FAIL: Correct 'Q8', got 'S8'
230 unexpected FAIL: Correct '!E8', got 'E8'
300 unexpected PASS!
360 unexpected PASS!
= 1.80 1.80 1.40 1.80 1.80 1.80

1070 unexpected FAIL: Correct 'A2', got 'J2'
= 1.80 1.80 1.80 2.00 1.80 1.80

1070 unexpected FAIL: Correct 'A2', got 'J2'
= 1.80 1.80 1.80 2.20 1.80 1.80

1070 unexpected FAIL: Correct 'A2', got 'J2'
= 1.80 1.80 1.80 2.00 1.80 1.80

= 1.80 1.80 1.80 1.80 2.00 1.80

70 unexpected PASS!
= 1.80 1.80 1.80 1.60 1.80 1.80

50 unexpected FAIL: Correct 'O10', got 'N17'
70 unexpected PASS!
160 unexpected FAIL: Correct 'B14|G13', got 'O4'
= 1.80 1.80 1.80 1.40 1.80 1.80

= 1.80 1.80 1.80 1.80 1.60 1.80

= 1.80 1.80 1.80 1.80 1.40 1.80

= 1.80 1.80 1.80 1.80 1.20 1.80

= 1.80 1.80 1.80 1.80 1.00 1.80

= 1.80 1.80 1.80 1.80 0.80 1.80

= 1.80 1.80 1.80 1.80 0.00 1.80

130 unexpected FAIL: Correct 'M15|J15|N16|J16', got 'C15'
1020 unexpected PASS!
1050 unexpected FAIL: Correct 'M15|J15|N16|J16', got 'C15'
1160 unexpected PASS!
= 1.80 1.80 1.80 1.80 1.80 2.00

130 unexpected FAIL: Correct 'M15|J15|N16|J16', got 'C15'
1020 unexpected PASS!
1050 unexpected FAIL: Correct 'M15|J15|N16|J16', got 'C15'
1090 unexpected FAIL: Correct 'O5', got 'O17'
1160 unexpected PASS!
= 1.80 1.80 1.80 1.80 1.80 2.20

80 unexpected FAIL: Correct 'G6', got 'F7'
140 unexpected FAIL: Correct 'P14', got 'H17'
440 unexpected FAIL: Correct '!T3', got 'T3'
= 1.80 1.80 1.80 1.80 1.80 1.60

Dan