Ticket #159 (new defect)

Opened 19 months ago

Last modified 19 months ago

Endgame

Reported by: sanxiyn Owned by: gnugo
Priority: normal Milestone: 3.8
Component: source Version:
Severity: normal Keywords:
Cc: patch: no

Description

gnugo plays A8 and lose. It should play A7!

Attachments

endgame.sgf (272 bytes) - added by sanxiyn 19 months ago.

Regression Results

Attachment Rev. PASS FAIL Nodes Status

Change History

Changed 19 months ago by sanxiyn

Changed 19 months ago by nando

The move at A8 is overvalued, because

  • per pattern EE401, it is considered a strategic defense of B8
  • the dragon B8 is considered a tiny bit weak, or said differently, it is seen as not completely safe: "only" 2 certain eyes, where the weakness computation (see crude_dragon_weakness() and the genus2weakness[] array in dragon.c) only considers 3 eyes or more as 100% safe.

Possible solutions:

  • remove the d class from all endgame patterns, on the basis that those usually lack enough context information to make such a call (on the other hand, someone probably had a reason to put those there...)
  • improve weakness and/or genus evaluation (clearly, that's easier said than done)

Sanxiyn, is it difficult for you to group those problems and submit them as patches for the regression suite ?

Note: See TracTickets for help on using tickets.