[Top][All Lists]

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

RE: [gnugo-devel] GNU Go 3.1.9

From: Måns Ullerstam
Subject: RE: [gnugo-devel] GNU Go 3.1.9
Date: Mon, 8 Oct 2001 08:27:46 +0200

Thanks Dan for going through my patterns, I would like to know what happened
to my 9x9 database. It was a bit big to go over by hand, pattern by pattern
(3000 patterns). Let me send you a new database for 9x9 with less patterns
and raise the standard of that one.

The tests with shdaing out parts of the fullboard patterns with question
marks is moving along. I now have a version of extract_fuseki that handles
this. I am unsure of the results of this though, I haven't tried it in
actual play.

It is interesting, but might lead to the case where we rewrite more of
gnugo. After I have shaded parts of the board, the patterns will match even
later on in the game. Not to confuse other parts of the system, I still
believe we should use these automatically generated patterns for the first
moves only and keep the value of 75.

-----Original Message-----
From: address@hidden [mailto:address@hidden
Behalf Of Daniel Bump
Sent: den 8 oktober 2001 05:14
To: address@hidden
Cc: address@hidden
Subject: [gnugo-devel] GNU Go 3.1.9

GNU Go 3.1.9 is now available at:

The CVS release tag is rel-3-1-9.

As with 3.1.8, regression/BREAKAGE shows that Gunnar and
Trevor have been busy. (Myself, I fixed one or two
things but haven't had much time to work on the program
this week.)

Watching games on NNGS I have the subjective impression
that the program is playing well.  Around Monday I ran
100 games of 3.1.8 against 3.0.0 and 3.1.8 won 55. This
could well be a statistical fluxuation but it will be
interesting to see if 3.1.9 does as well or better.

The following patches are in:

gunnar_1_9.6. Tuning, etc.
tristan_1_9.2. Reads connections.
teun_1_9.4. Point orientation.
inge_1_9.3. Makes eye_data one-dimensional.
gunnar_1_9.5. Fixes some of Trevor's regressions.
trevor_1_9.1. Needed to trap errors on Win32.
trevor_1_9.2. Adds informational comments to trevor.tst.
trevor_1_9.3. Adds two files to regression/
teun_1_9.2. Completes the -i and -o mkpat options.
trevor_reg_1_9.tar.gz. A new pile of regressions.
gunnar_1_9.4. More 1D stuff, mostly in worm.c.
gunnar_1_9.3. Minor bugfix.
gunnar_1_9.2. Revises compute_primary_domains().
vcrepair_1_9.2. CR/LF madness fixed in .dsp files.
gunnar_1_9.1. Lots of one-dimensional conversion and cleanup.
teun_1_9.1. Adds -i and -o options to mkpat.
worm_stones_1_9.1. GTP worm_stones OK when stackp>0, can take color
inge_1_9.2.gz. Big patch making dragon[][] and worm[][] one-dimensional.
vcrepair_1_9.1. Remember to compile fuseki.c under VC.

There were a couple of other patches that went directly into the
CVS and were documented in the devel list.

The principal omission is Man's revised fuseki.db. I am slowly going
through these 1500 patterns and deleting a few that are obviously
bad. When I am done this will probably be accepted. I'm up to pattern 223.

- revised examine_move_safety heuristics
- owl_determine_life() revised
- new gtp functions connect and disconnect
- naive_ladder() made global
- warnings eliminated from readconnect.c
- most functions in readconnect.c made static
- add point reorientation functions
- add some comments to dfa.c
- bugfix and cleanup in naive_ladder()
- new interface/GoImage/
- struct eye_data arrays changed from 2D to 1D
- 9x9 case revised in fuseki()
- bugfix in goal detection in matchpat, don't accept empty goal vertices
- bugfix in gtp_genmove()
- new test cases in owl.tst
- further 1D revisions in worm.c
- find_lunch() moved from utils.c to worm.c
- is_same_worm() converted to 1D
- bugfix in list_move_reasons()
- revised test case in niki.tst
- tuning
- owl tuning
- connection tuning
- new implementation of compute_primary_domains()
- new eye pattern
- bugfix in fuseki.c
- cleanup and further 1D conversion in optics.c
- sgf_decideeye doesn't require owl computation
- 1D revisions in worm.c
- is_worm_origin(), propagate_worms(), does_attack(), does_defend() are 1D
- NNGS customization removed from ascii interface
- gtp_worm_stones can be run with stackp>0 and can take a color argument
- mkpat can take -i and -o options
- MIN_BOARD is 5


gnugo-devel mailing list

reply via email to

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