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
9 To build Pachi, simply type:
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.
31 The default engine plays by Chinese rules and should be about
32 3d KGS strength on 9x9. On 19x19, it might be about KGS 2k, 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 1d rank.
35 When using a large cluster (64 machines, 20 cores each), it maintains
36 KGS 3d and has won 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.
44 At the same time, we keep trying a wide variety of other approaches
45 and enhancements. Pachi is an active research platform and quite a few
46 improvements have been already achieved. We rigorously play-test new
47 features and enable them by default only when they give a universal
51 By default, Pachi will run on a single CPU core, taking up to 1.4GiB
52 of memory, not pondering and taking completely arbitrary amount of time
53 per turn. You can adjust these parameters by passing it extra command
56 ./pachi -t _1200 threads=8,max_tree_size=3072,pondering
58 This will make Pachi play with time settings 20:00 S.D. (unless it
59 gets told otherwise over GTP), with 8 threads, taking up to 3GiB
60 of memory (+ several tens MiB as a constant overhead) and thinking
61 during the opponent's turn as well.
63 For now, there is no comprehensive documentation of options, but you
64 can get a pretty good idea by looking at the uct_state_init() function
65 in uct/uct.c - you will find the list of UCT engine options there, each
66 with a description. At any rate, usually the three options above are
67 the only ones you really want to tweak.
70 Except UCT, Pachi supports a simple idiotbot-like engine and an example
71 treeless MonteCarlo-player. The MonteCarlo simulation ("playout")
72 policies are also pluggable, by default we use the one that makes use of
73 heavy domain knowledge.
75 Other special engines are also provided:
76 * a "distributed" engine for cluster play; the description at the top of
77 distributed/distributed.c should provide all the guidance
78 * a simple "replay" engine that will simply play moves according
79 to the playout policy suggestions
80 * few other purely for development usage
83 Pachi can be used as a test opponent for development of other go-playing
84 programs. For example, to get the "plainest UCT" player, use:
86 ./pachi -t =5000 policy=ucb1,playout=light,prior=eqex=0,dynkomi=none
88 This will fix the number of playouts per move to 5000, switch the node
89 selection policy from ucb1amaf to ucb1 (i.e. disable RAVE), switch the
90 playouts from heuristic-heavy moggy to uniformly random light, stop
91 prioring the node values heuristically, and turn off dynamic komi.
93 You can of course selectively re-enable various features or tweak this
94 further. But please note that using Pachi in this mode is not tested
95 extensively, so check its performance in whatever version you test
96 before you use it as a reference.
98 Note that even in this "basic UCT" mode, Pachi optimizes tree search
99 by considering board symmetries at the beginning. Currently, there's no
100 easy option to turn that off. The easiest way is to tweak board.c so
101 that board_symmetry_update() has goto break_symmetry at the beginning
102 and board_clear has board->symmetry.type = SYM_NONE.
108 Pachi can also help you analyze your games by being able to provide
109 its opinion on various positions. The user interface is very rudimentary,
110 but the ability is certainly there.
112 There are currently two Pachi interfaces provided for this purpose. One
113 possibility is to evaluate all moves within a given game and show how
114 the winrates for both players evolved - i.e. who was winning at which
115 game stage. This is implemented using the `tools/sgf-analyse.pl` script.
116 See the comment on top of the script about its usage.
118 Alternatively, Pachi can evaluate all available moves in a given situation
119 and for each give a value between 0 and 1 representing perceived
120 likelihood of winning the game if one would play that move. I.e. it can
121 suggest which moves would be good and bad in a single given situation.
123 To achieve the latter, follow few simple steps:
125 1. Take an SGF record of your game and generate a GTP file from it:
127 tools/sgf2gtp.pl <file.sgf >file.gtp
129 2. Open the file.gtp in your favorite plaintext editor and trim it after
130 (remove all lines following) the move producing the situation you wish
133 3. Instead, add the following line at the bottom of file.gtp:
137 (replace `black` with `white` if it is white to play).
139 4. Run Pachi as follows:
141 ./pachi -t =500 -d 0 <file.gtp | sed -n '/^=0/,${s/^=0 //;p}'
143 If you want to know more details on what is Pachi thinking about the
144 various moves, remove the `-d 0` and `| sed -n ...` parts. To improve
145 the accuracy of values (at the expense of run time), raise the value
146 of 500 (try 2000 or 10000; 100000 will usually already take couple of
147 hours). The values will be most useful in the middle game; in fuseki
148 and most yose situations, expect a lot of noise.
154 The aim of the software framework is to make it easy to plug your
155 engine to the common infrastructure and implement your ideas while
156 minimalizing the overhead of implementing the GTP, speed-optimized
157 board implementation, etc. Also, there are premade random playout
158 and UCT tree engines, so that you can directly tweak only particular
159 policies. The infrastructure is pretty fast and it should be quite
160 easy for you (or us) to extend it to provide more facilities for
163 See the HACKING file for a more detailed developer's view of Pachi.
169 Pachi is distributed under the GPLv2 licence (see the COPYING file for
170 details and full text of the licence); you are welcome to tweak it as
171 you wish (contributing back upstream is welcome) and distribute
172 it freely, but only together with the source code. You are welcome
173 to make private modifications to the code (e.g. try new algorithms and
174 approaches), use them internally or even to have your bot play on the
175 internet and enter competitions, but as soon as you want to release it
176 to the public, you need to release the source code as well.
178 One exception is the Autotest framework, which is licenced under the
179 terms of the MIT licence (close to public domain) - you are free to
180 use it any way you wish.