gnugo-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[gnugo-devel] trevor_1_15.1


From: Trevor Morris
Subject: [gnugo-devel] trevor_1_15.1
Date: Tue, 20 Nov 2001 11:20:18 -0500

http://www.public32.com/games/go/trevor_1_15.1.tar.gz
 - regress.pl now accepts "problem sets"
 - regress.plx includes "unexpected result" list.
 - removed tactical reading move pass from owl.c
 - tuning.
 - a few test revisions.

The tactial reading pass removal causes about 10 unexpected failures
and 10 unexpected passes.  I corrected a few failures with simple
tuning.  The rest don't look serious, and the number of useless owl
reading nodes dropped by just over 20% (18362 to 14529).  It seems
like the owl patterns are mature enough to warrant this move now.

I tested GNU GO against some problems in "All About Life and Death"
volume one by Cho Chikun.  Most of the tuning in this patch is
based on those failures.  I can't provide the problems sets, due
to copyright concerns, however.  So, if you see references to,
for example cho1:161, that's a reference to this test file.


One last item of note:
  regress.pl now takes a "problem set", and runs only the problems
listed.  For example, here's a file specifying a problem set:

The following file lists all the failed test cases caused by removing
the tactical reading pass from owl.c.  Given to regress.pl, currently
it will only process trevor:590.  Remove the line "DONE", and it will
run all the listed test cases.  I've found this feature handy to keep
track of the problems that I'm looking at, and see how code changes
affect the set.

---ps.txt---

trevor:590
DONE

buzco:6    FAILED   Corrected problem.
global:4   FAILED   F5 now owl captures E4.  Hard to deny. Overvalued?
neurogo:3  FAILED  Q3 is huge - not a big problem.
nicklas1:1408 FAILED  GNU Go was right for the wrong reasons before.
nicklas1:1601 FAILED  old GNU Go missed the point that D7 was a cut.  Right
for wrong reason.
  # In particular here's the command that's different on 1601, but it
misses the point.
  # gg1118 --quiet -l games/nicklas/nicklas16.sgf -L 23 --decide-dragon C8
nicklas2:1407 FAILED  Old reading was right for wrong reason.
  #  gg1118 -l games/nicklas/nicklas14.sgf -L 61 --decide-dragon B8 -o x.sgf
owl:157     FAILED   !!!Not sure how best to fix this one.
owl:166     FAILED   !!!Not sure how best to fix this one.
owl_rot:150 FAILED   Now same failure as non-rot version!
strategy:27 FAILED   !!!Not sure how best to fix this one.
strategy:34 FAILED   Tough to prove that B move doesn't kill W.
viking:3    FAILED   !!!Not clear what happened here.





reply via email to

[Prev in Thread] Current Thread [Next in Thread]