| blob | 764b85c193ff62b05f1a89f0f1b5b547cf6295db |
1 Pachi can refer to: a simple modular framework for programs playing
2 the game of Go/Weiqi/Baduk, and a reasonably strong engine built
3 within this framework.
6 Installation
7 ------------
9 To build Pachi, simply type:
11 make
13 The resulting binary program `pachi` is a GTP client; connect to it
14 with your favorite Go program interface (e.g. gogui or qgo), or use
15 kgsGtp to connect it to KGS. (DO NOT make the GTP interface accessible
16 directly to untrusted users since the parser is not secure - see the
17 HACKING file for details.)
19 The pachi program can take many parameters, as well as the particular
20 engine being used; the defaults should be fine for initial usage,
21 see below for some more tips.
23 In case you hit compilation issues (e.g. when building on MacOS/X)
24 or want to change the build configuration, check the user configurable
25 section at the top of the Makefile.
28 Engine
29 ------
31 The default engine plays by Chinese rules and should be about
32 7d KGS strength on 9x9. On 19x19, it might be about KGS 1k, assuming
33 reasonable hardware, e.g. two-core Athlon64 machine. On a higher-end
34 (e.g. six-way Intel i7) machine, it can hold a solid KGS 2d rank.
35 When using a large cluster (64 machines, 20 cores each), it maintains
36 KGS 4d and has won e.g. a 7-stone handicap game against Zhou Junxun 9p.
38 By default, Pachi currently uses the UCT engine that combines
39 Monte Carlo approach with tree search; UCB1AMAF tree policy using
40 the RAVE method is used for tree search, while the Moggy playout
41 policy using 3x3 patterns and various tactical checks is used for
42 the semi-random Monte Carlo playouts. Large-scale board patterns
43 are used in the tree search.
45 At the same time, we keep trying a wide variety of other approaches
46 and enhancements. Pachi is an active research platform and quite a few
47 improvements have been already achieved. We rigorously play-test new
48 features and enable them by default only when they give a universal
49 strength boost.
51 How to run
52 ~~~~~~~~~~
54 By default, Pachi will run on a single CPU core, taking up to 1.4GiB
55 of memory and taking a little under 15 seconds per move. You can
56 adjust these parameters by passing it extra command line options:
58 ./pachi -t _1200 threads=8,max_tree_size=3072
60 This will make Pachi play with time settings 20:00 S.D. with 8 threads,
61 taking up to 3GiB of memory (+ several tens MiB as a constant overhead)
62 and thinking during the opponent's turn as well.
64 Pachi can use an opening book in a Fuego-compatible format - you can
65 obtain one at http://gnugo.baduk.org/fuegoob.htm and use it in Pachi
66 with the -f parameter:
68 ./pachi -f book.dat ...
70 You may wish to append some custom Pachi opening book lines to book.dat;
71 take them from the book.dat.extra file. If using the default Fuego book,
72 you may want to remove the lines listed in book.dat.bad.
74 Pachi can also use a pattern database to improve its playing performance.
75 You can get it at http://pachi.or.cz/pat/ - you will also find further
76 instructions there.
78 For now, there is no comprehensive documentation of options, but you
79 can get a pretty good idea by looking at the uct_state_init() function
80 in uct/uct.c - you will find the list of UCT engine options there, each
81 with a description. At any rate, usually the three options above are
82 the only ones you really want to tweak.
84 Greedy Pachi
85 ~~~~~~~~~~~~
87 Normally, Pachi cares only for win or loss and does not take into
88 account the point amount. This means that it will play slack endgame
89 when winning and crazy moves followed with a resign when losing.
91 It may give you a more pleasurable playing experience if Pachi
92 _does_ take into account the point size, strives for a maximum
93 (reasonable) win margin when winning and minimal point loss when
94 losing. This is possible by using the maximize_score parameter, e.g.:
96 ./pachi -t _1200 threads=8,maximize_score
98 This enables an aggressive dynamic komi usage and end result margin
99 is included in node values aside of winrate. Pachi will also enter
100 scoring even when losing (normally, Pachi will never pass in that case).
101 Note that if you pass any 'dynkomi' parameter to Pachi, you will reset
102 the values set by 'maximize_score'.
104 Note that Pachi in this mode may be slightly weaker, and result margin
105 should not be taken into account when judging either player's strength.
106 During the game, the winning/losing margin can be approximated from
107 Pachi's "extra komi" or "xkomi" reporting in the progress messages.
109 Experiments and Testing
110 ~~~~~~~~~~~~~~~~~~~~~~~
112 Except UCT, Pachi supports a simple idiotbot-like engine and an example
113 treeless MonteCarlo-player. The MonteCarlo simulation ("playout")
114 policies are also pluggable, by default we use the one that makes use of
115 heavy domain knowledge.
117 Other special engines are also provided:
118 * a "distributed" engine for cluster play; the description at the top of
119 distributed/distributed.c should provide all the guidance
120 * a simple "replay" engine that will simply play moves according
121 to the playout policy suggestions
122 * a simple "patternplay" engine that will play moves according to the
123 learned patterns
124 * few other purely for development usage
126 Pachi can be used as a test opponent for development of other go-playing
127 programs. For example, to get the "plainest UCT" player, use:
129 ./pachi -t =5000 policy=ucb1,playout=light,prior=eqex=0,dynkomi=none,pondering=0,pass_all_alive
131 This will fix the number of playouts per move to 5000, switch the node
132 selection policy from ucb1amaf to ucb1 (i.e. disable RAVE), switch the
133 playouts from heuristic-heavy moggy to uniformly random light, stop
134 prioring the node values heuristically, turn off dynamic komi, disable
135 thinking on the opponent's time and make sure Pachi passes only when
136 just 10% alive stones remain on the board (to avoid disputes during
137 counting).
139 You can of course selectively re-enable various features or tweak this
140 further. But please note that using Pachi in this mode is not tested
141 extensively, so check its performance in whatever version you test
142 before you use it as a reference.
144 Note that even in this "basic UCT" mode, Pachi optimizes tree search
145 by considering board symmetries at the beginning. Currently, there's no
146 easy option to turn that off. The easiest way is to tweak board.c so
147 that board_symmetry_update() has goto break_symmetry at the beginning
148 and board_clear has board->symmetry.type = SYM_NONE.
151 Analysis
152 --------
154 Pachi can also help you analyze your games by being able to provide
155 its opinion on various positions. The user interface is very rudimentary,
156 but the ability is certainly there.
158 There are currently several Pachi interfaces provided for this purpose.
160 Winrate Development
161 ~~~~~~~~~~~~~~~~~~~
163 Pachi can evaluate all moves within a given game and show how
164 the winrates for both players evolved - i.e. who was winning at which
165 game stage. This is implemented using the `tools/sgf-analyse.pl` script.
166 See the comment on top of the script about its usage.
168 Move Ranking
169 ~~~~~~~~~~~~
171 Pachi can evaluate all available moves in a given situation
172 and for each give a value between 0 and 1 representing perceived
173 likelihood of winning the game if one would play that move. I.e. it can
174 suggest which moves would be good and bad in a single given situation.
176 To achieve the latter, note the number of move at the situation you
177 want to evaluate and run the `tools/sgf-ratemove.sh` script.
178 See the comment on top of the script about its usage.
180 Pattern Move Hinting
181 ~~~~~~~~~~~~~~~~~~~~
183 Pachi can show instantenous pattern-based move suggestions very much
184 like for example Moyo Go Studio (though of course without a GUI).
185 You can use the Move Ranking method above (tools/sgf-ratemove.sh),
186 but pass it an extra parameter '-e patternplay'.
189 Framework
190 ---------
192 The aim of the software framework is to make it easy to plug your
193 engine to the common infrastructure and implement your ideas while
194 minimalizing the overhead of implementing the GTP, speed-optimized
195 board implementation, etc. Also, there are premade random playout
196 and UCT tree engines, so that you can directly tweak only particular
197 policies. The infrastructure is pretty fast and it should be quite
198 easy for you (or us) to extend it to provide more facilities for
199 your engine.
201 See the HACKING file for a more detailed developer's view of Pachi.
203 Also, if you are interested about Pachi's architecture, algorithms
204 etc., consider taking a look at Petr Baudis' Master's Thesis:
206 http://pasky.or.cz/go/prace.pdf
209 Licence
210 -------
212 Pachi is distributed under the GPLv2 licence (see the COPYING file for
213 details and full text of the licence); you are welcome to tweak it as
214 you wish (contributing back upstream is welcome) and distribute
215 it freely, but only together with the source code. You are welcome
216 to make private modifications to the code (e.g. try new algorithms and
217 approaches), use them internally or even to have your bot play on the
218 internet and enter competitions, but as soon as you want to release it
219 to the public, you need to release the source code as well.
221 One exception is the Autotest framework, which is licenced under the
222 terms of the MIT licence (close to public domain) - you are free to
223 use it any way you wish.