Ticket #90 (closed enhancement: fixed)

Opened 6 years ago

Last modified 6 years ago

New dragon data field strategic_size to estimate strategical effects

Reported by: arend Owned by: arend
Priority: normal Milestone: 3.7.9
Component: source Version:
Severity: normal Keywords:
Cc: patch: yes

Description

This patch adds a new field .strategic_size in the dragon_data2. It consists of the effective_size, plus a bonus for all weak neighboring dragons belonging to opponent.

.strategic_size is then used in place of .effective_size for the computation of strategic effects during move valuation. 

arion:1         PASS Q6 [Q6]            Very good
arion:3         PASS J3 [J3]            Good but lucky
viking:3        PASS M15 [M15|B7]       Very good
trevorb:290     PASS E10 [!E13]         Random
strategy2:53    FAIL E10 [S15]          Somewhat random, owl misread on left
             side now causes consequent valuation of E10
strategy2:79    PASS N3 [N3]            Valuation of N3 improved, but
             S4 (played by CVS) isn't bad either
trevor:220      FAIL A7 [E8]            Bad, A7 overvalued.
buzco:4         FAIL E15 [D13]          Acceptable.
nngs:230        PASS D6 [E9|G7|H2|K6|N5|D6]     Good.
nngs:1670       PASS C18 [C18]          Good
strategy3:119   PASS D9 [D9|J3]         Good (if a little lucky)
strategy4:153   FAIL O11 [Q9]           ok (Q9 has Q15/R10 cut reason missing)
strategy4:166   FAIL D8 [D11]           Bad, but mostly semeai mistake
handtalk:11     PASS C11 [C11|D10]      Very good.
nngs3:140       PASS D9 [D9]            Very good.
nngs3:470       PASS P11 [!J17]         Good.
nngs3:480       PASS N14 [N14|O14]      Good.
nngs3:490       FAIL N14 [P15]          Quite bad.
nngs3:710       FAIL C2 [Q2]            Bad but ok.
nngs4:560       FAIL F7 [B13]           I don't understand this test.
nngs4:990       PASS R15 [R16|R15|Q15]  Very good.
nngs4:1050      FAIL Q8 [O9|O10]        Ok.
century2002:60  FAIL F10 [Q14]          Good.
safety:5        PASS K8 [K8|L9|L7]      Good.
ninestones:370  FAIL E12 [R5]           Bad.
ninestones:490  PASS R10 [R10]          Good (lucky)
arend2:30       PASS K3 [K3|J4|M5]      Very good.
13x13b:14       PASS J9 [J9]            Good.
seki:803        PASS C2 [C2]            I don't understand this test.
9x9:520         PASS B2 [E5|D5|B2|E3|F3|F4]     Good.
kgs:230         FAIL O11 [J12|J11|L10|L9|J19]   Very bad.
kgs:400         PASS P8 [P8|P7|Q13]     Good.
Total nodes: 1684329924 3284739 12473414
Total uncertainty: 47.86
22 PASS
12 FAIL

Overall, the breakage looks very good. It causes GNU Go to play urgent moves more often. If there is any pattern in the FAILs, it is that strategic move are now occasionally a little overvalued compared to owl attack/defenses. Presumable, this could be solved by adding a strategic bonus for owl moves depending on weakness of the neighbors.

Attachments

arend_7_9.12.diff Download (8.3 KB) - added by arend 6 years ago.

Regression Results

Attachment Rev. PASS FAIL Nodes Status
arend_7_9.12.diff Download never tested

Change History

Changed 6 years ago by arend

comment:1 Changed 6 years ago by arend

  • Owner changed from gnugo to arend

comment:2 Changed 6 years ago by arend

  • Status changed from new to closed
  • Resolution set to fixed

comment:3 Changed 6 years ago by gunnar

Regarding seki:803, see Davies' book about Life and Death, page 125, diagram 7.

Note: See TracTickets for help on using tickets.