Time for a refreshing reading patch. This adds the static function squeeze_moves() in reading.c, which looks for moves to squeeze out liberties from superstrings. Typical examples are reading:203,204.
- new static function squeeze_moves() in reading.c The breakage is (computed before the last patches went into CVS): reading:203 PASS 1 E9 [1 E9] filllib:35 FAIL A7 [A8|B8] trevorb:600 FAIL N7 [F2|F3] trevor:290 PASS G8 [!A9] nngs4:730 PASS C17 [C17] seki:206 PASS A1 [A1] seki:803 FAIL C1 [C2] seki:811 PASS A1 [A1] 5 PASS 3 FAIL Total nodes: 1675157568 3156143 13093899 (+0.19% -0.09% -0.16%) reading:203 Good pass, targeted testcase. trevor:290 Good pass. Now it's understood that white A9 can tactically be attacked with ko (previous result 0). nngs4:730 The difference is whether E17 can be tactically defended. Without the patch D17 is found as a defense, with the patch no defense is found. Unclear what's correct. seki:206 seki:811 Very good. Ko attack is now found on A4. (The two testcases are similar but by no means identical.) filllib:35 Accidental. The primary change is that black A5 is understood to no longer be tactically safe, which is very good. trevorb:600 With patch a bogus atari-atari defense appears at N7. The only change in reading is B:N7 W:N9 B:N11 W:M10 B:H13 W:K13 attack J13 (22) 0 PASS 1 M12 which of course is very good. Why the atari-atari reading makes a mistake is unclear. seki:803 Many changes in reading results: 1. B:A2 W:C1 B:B3 attack B3 (16) 0 PASS 1 A1 2. B:C2 W:C1 B:B1 attack A4 (16) 0 PASS 1 A3 3. W:A3 B:B1 defend A4 (16) 5 C2 5 C1 4. W:B1 B:B3 W:C2 B:A2 attack A2 (16) 0 PASS 1 D1 5. B:B1 attack A4 (17) 0 PASS 1 C1 6. B:C2 W:C1 B:B3 attack C1 (16) 0 PASS 1 B1 7. B:C2 W:C1 B:B3 W:A1 attack C1 (16) 0 PASS 1 A2 8. B:A2 W:C1 attack A4 (16) 0 PASS 1 A3 9. B:A2 W:C1 B:B3 attack B3 (16) 0 PASS 1 A1 10. W:C1 defend A4 (17) 5 A2 4 A2 11. W:A3 B:C1 W:C2 defend C3 (16) 5 B1 4 B1 12. W:A3 B:C1 W:C2 B:B1 attack A4 (17) 0 PASS 1 A1 13. W:A3 B:C2 W:C1 B:B1 defend A4 (16) 5 D1 4 D1 14. W:A3 B:B1 defend A4 (16) 5 C2 5 C1 15. W:C2 B:A2 W:D1 B:B1 attack B1 (16) 0 PASS 4 A3 16. B:A2 W:C2 B:B3 W:C1 attack D2 (16) 0 PASS 1 A1 Number 1,3,4,5,8,9,10,11,12,14,15,16 good. Number 2,6,7 bad. 13 bad but probably not the fault of the patch. Overall an improvement. /Gunnar _______________________________________________ gnugo-devel mailing list [email protected] http://lists.gnu.org/mailman/listinfo/gnugo-devel
