[PATCH] board_handicap(): Correct placement for three-stone handicaps
Merge branch 'master' of ssh://repo.or.cz/srv/git/pachi
ucb1amaf_update criticality stats: Fix invalid memory access on pass
board_clear(): Cache set up board structures - major speedup for loading fbook
board_clear(): Factor out board_init_data()
fbook: Enable debug prints
fbook_init(): Return NULL if the book turned out to stay empty
This also prevents cache reset on such occassion.
This also prevents cache reset on such occassion.
fbook caching: Store last loaded fbook and reuse it if it matches current board
fbook: Increase fbook_hash_bits to 20 to accomodate even large fbooks like ra6
Merge branch '2lib'
can_atari_group(): When switching to selfatari cousin, check defense_is_hopeless()
selfatari_cousin(): Make info on by-group available
can_atari_group(): Factor out hope check to defense_is_hopeless()
can_atari_group(): Add massive amount of debugging prints
TODO: Liberty maps (developed in libmap branch), reverse status learning
tools/sgf2gtp.pl: Support HA[] sgf tag
tools/sgf2gtp.pl: Accept filename as optional argument
gtp: Add fixed_handicap to list of supported commands
Pachi Hakugen-devel 8.99
Pachi Shuhaku 8.00
Parallel make: Fix pachi being built before some libraries
fbook: Share no-handicap fbook even for low komi values
Suggested by jlg.
Suggested by jlg.
fbook_init(): do not force h1 for large negative komi
Dynamic komi: fix linear_permove() when pondering
Criticality: New defaults based on tuning by jlg
Moggy ladderrate: Default to 80, confirmed by jlg
Fix parsing of mqprob, moves & handicap_value parameters
Merge branch 'master' of ssh://repo.or.cz/srv/git/pachi
Add final_score to known_commands
Implement known_command; make gtp_is_valid() stricter
Makefile.lib: Allow parallel build
fbook: Add support for handicap lines; do not use normal lines in handicap games
Moggy: cap_stone_denom=200, ladderrate=40
This turns out the capture stone weighting and preferring ataris that
walk the ladder. I have measured both to be gains with reasonably high
probability, but more tuning is needed to find the optimum.
This turns out the capture stone weighting and preferring ataris that
walk the ladder. I have measured both to be gains with reasonably high
probability, but more tuning is needed to find the optimum.
UCB1AMAF crit_*: New defaults based on very rough tuning
On 19x19, this may be as much as 50 ELO gain vs. crit_rave=0. More
tuning still going on.
On 19x19, this may be as much as 50 ELO gain vs. crit_rave=0. More
tuning still going on.
Merge branch 'master' into criticality
Conflicts:
uct/tree.c
Conflicts:
uct/tree.c
tree_node_dump(): Much more condensed output
UCB1AMAF: Add criticality support, with many tuning options
crit_rave=2, crit_min_playouts=32, crit_negative=1, crit_amaf=1
crit_rave=2, crit_min_playouts=32, crit_negative=1, crit_amaf=1
tree_node_criticality(): Add further explanation for drunk readers like me
Merge branch 'master' into criticality
tree_node_criticality(): Return floating_t instead of double
Moggy: Turn off laderrate for now
Optimum value not definitely determined yet, but it is likely that the
gain is negative for -1.
Optimum value not definitely determined yet, but it is likely that the
gain is negative for -1.
Merge branch 'ladders'
README: Wording tweak
media: Add two pictures of Pachi; let's not neglect its personality!
TODO: Remove two stale items
Rename the main program binary from zzgo to pachi
'zzgo' is still residuum from the earliest times when I was not decided
on the program name. Good bye!
'zzgo' is still residuum from the earliest times when I was not decided
on the program name. Good bye!
Header guards: Rename ZZGO_* to PACHI_*
UCT max_tree_size: Lower the default to 1.4GiB; adjust and polish README
Many notebooks still sold have just 2GiB of memory, plus some memory may
be taken by OS and other apps. Overally, the 1.4GiB limit is IMHO a much
safer default, especially given there is no pondering and just single
core used by default too.
Many notebooks still sold have just 2GiB of memory, plus some memory may
be taken by OS and other apps. Overally, the 1.4GiB limit is IMHO a much
safer default, especially given there is no pondering and just single
core used by default too.
UCT fast_alloc: Make default
README: Move the installation instructions to the top
Deprecate the zzgo-profiled make target
README: Grammar polishing, rewording and extending the text
More user friendly build-time configuration
The Makefile now contains descriptions for all available build time
options.
The Makefile now contains descriptions for all available build time
options.
Moggy stone weighing: Fix case when cap_stone_denom == 0
Moggy 1lib prior: Add support for stone weighing, turned off by default
New options cap_stone_min, cap_stone_max, cap_stone_denom. Needs a lot of tuning.
New options cap_stone_min, cap_stone_max, cap_stone_denom. Needs a lot of tuning.
Merge branch 'master' into ladders
13 years agoMoggy pattern3, group_atari_check(): Ignore ladder extensions in case counter-capture...
Moggy pattern3, group_atari_check(): Ignore ladder extensions in case counter-capture is possible
can_countercapture(): Export interface
Merge branch 'master' into ladders
Merge branch 'master' of git+ssh://repo.or.cz/srv/git/pachi
Moggy: Play the proper atari in case we can catch a group in ladder
middle_ladder_walk(): Fix not noticing ladders that skim corners at the edges
is_middle_ladder(): Factor out middle_ladder_walk()
Distributed engine: Wait for only 3/4 slaves and at most 0.5s for fast commands
This reduces the risk of losing on time when replaying a whole game after KGS disconnect.
This reduces the risk of losing on time when replaying a whole game after KGS disconnect.
UCT local_tree_depth_decay: Apply based on start of sub-sequence, not end of descent
uct_playout(): Remove superfluous depth variable
Autotest: Fix loading rc file when /bin/sh is not bash
Pachi Shuhaku-devel 7.99
Pachi Chihaku 7.00
group_atari_check(): Do not print misleading 'no ladder' messages
tree_node_dump(): Include criticality
Support undo for pass moves
This form of undo is required by KGS to settle disputes on dead groups.
It is needed in particular for the distributed engine because replaying
the entire game is too slow and causes loss on time.
This form of undo is required by KGS to settle disputes on dead groups.
It is needed in particular for the distributed engine because replaying
the entire game is too slow and causes loss on time.
board_effective_handicap(): Take komi into account when no handicap
This is a fix for 3679219, pointed out by pasky.
This is a fix for 3679219, pointed out by pasky.
Linear dynkomi: Allow different parameters for black and white
This allows using linear dynkomi also for white. Otherwise
white resigns immediately in high handicap games.
This allows using linear dynkomi also for white. Otherwise
white resigns immediately in high handicap games.
board_effective_handicap(): Always return 0 for even games
Merge branch 'master' of git://repo.or.cz/pachi into atari
UCT: Track criticality statistics
This just adds all the book-keeping. We do not use the criticality
information anywhere yet.
This just adds all the book-keeping. We do not use the criticality
information anywhere yet.
UCT: Introduce tree_node_parity() macro
can_atari_group(): Remove "lumpy" counter-example which degrades performance
This partially reverts 66735b2, keeping the good change owner->to_play.
This partially reverts 66735b2, keeping the good change owner->to_play.
group_atari_check(): Do not consider any defending move when attacking
This avoids expensive checks when attacking and simplifies
the situation (capture laddered group). Needs lcapturerate=90
for optimal results.
This avoids expensive checks when attacking and simplifies
the situation (capture laddered group). Needs lcapturerate=90
for optimal results.
UCT local_tree_pseqroot: Remove unused setting
Move some helper scripts and files to tools/ subdirectory
Remove the treepool feature
The treepool has been a failed experiment so far, and it does not seem
likely we will get to make it a real playing strength contribution
anytime soon. Therefore, let's leave it just in the git history.
I have kept the 'significant node' feature since it is used also in some
unmerged branches, is a small piece of code and can come useful for
other stuff I have in mind.
The treepool has been a failed experiment so far, and it does not seem
likely we will get to make it a real playing strength contribution
anytime soon. Therefore, let's leave it just in the git history.
I have kept the 'significant node' feature since it is used also in some
unmerged branches, is a small piece of code and can come useful for
other stuff I have in mind.
Moggy local_2lib_check(): Avoid duplicate 2-lib checks
In case of 2-libs, these were occuring a lot and are rather expensive.
In case of 2-libs, these were occuring a lot and are rather expensive.
Moggy: Add support for specifying MQ_NAKADE gamma
Moggy nakaderate: Introduce, try nakade around last move
Tactics nakade_point(): Simple eye-piercing tool
Selfatari setup_nakade_or_snapback(): Slim down the group loop
Use the group iterator just to find out the other contended liberty, do
all the other tests outside. If there is no common other liberty, that
should mean this is not a true nakade.
Use the group iterator just to find out the other contended liberty, do
all the other tests outside. If there is no common other liberty, that
should mean this is not a true nakade.
UCT: Add basic support for graceful jigo handling (integer komi games)
check_group_atari(): If we cannot take liberty due to selfatari, try connecting first
can_atari_group(): Fix the bulkiness check not to prevent connections
check_group_atari(): Pass correct owner parameter
This might help fix use_def_no_hopeless.
This might help fix use_def_no_hopeless.
Selfatari setup_nakade_or_snapback(): Discern nakade and seki properly
We do not actually examine the shape, but instead in cooperation with
Brett Combs, we have devised a set of static rules that should
characterize the proper bulkiness with almost 100% reliability.
We do not actually examine the shape, but instead in cooperation with
Brett Combs, we have devised a set of static rules that should
characterize the proper bulkiness with almost 100% reliability.
Selfatari: Track neighbors corresponding to groups in groupneis[]
Board: Introduce simple group_stone_count() utility
Selfatari: Check groupids[] against duplicates
Moggy local_nlib_check(): Also look at diagonal neighbors
This is wise to do since we may have connected a false eye.
This is wise to do since we may have connected a false eye.
Moggy: Better nlibrate default for 19x19