Ticket #184 (new defect)

Opened 20 months ago

False and half eye detection at the edge

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

Description

The mailing list thread starting with http://lists.gnu.org/archive/html/gnugo-devel/2007-03/msg00011.html discusses weaknesses in the detection of false and half eyes at the edge.

Specifically, in this position





the bigger white eyespace gets the graph "...!" which implies a secure eye. More properly the graph should be ".h!". The immediate problem is that topological analysis only is applied for non-marginal eyespace points with at most one neighbor inside the eyespace.

Attachments

testcases_7_11.1.diff (0.8 kB) - added by gunnar 20 months ago.
Four testcases from two positions.
optics_7_11.1.diff (0.5 kB) - added by gunnar 20 months ago.
Proposed optics revision from http://lists.gnu.org/archive/html/gnugo-devel/2007-04/msg00004.html
optics_7_11.1b.diff (0.9 kB) - added by gunnar 20 months ago.
Alternative optics revision.

Regression Results

Attachment Rev. PASS FAIL Nodes Status
optics_7_11.1.diff 2383 28 40 +1.2% +1.4% +0.37% details
optics_7_11.1b.diff 2383 26 34 +0.76% +0.88% +0.26% details
testcases_7_11.1.diff never tested

Change History

Changed 20 months ago by gunnar

Four testcases from two positions.

Changed 20 months ago by gunnar

Proposed optics revision from http://lists.gnu.org/archive/html/gnugo-devel/2007-04/msg00004.html

Changed 20 months ago by gunnar

Alternative optics revision.

Note: See TracTickets for help on using tickets.