1 libDAI - A free/open source C++ library for Discrete Approximate Inference

3 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

5 Version: git commit fc04cce2b5027d58acfe965cee7ed01ec1dbb9a7

6 Date: Mon Nov 16 13:14:56 2009 +0100

7 See also: http://www.libdai.org

9 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

11 License

13 libDAI is free software; you can redistribute it and/or modify it under the

14 terms of the GNU General Public License as published by the Free Software

15 Foundation; either version 2 of the License, or (at your option) any later

16 version.

18 libDAI is distributed in the hope that it will be useful, but WITHOUT ANY

19 WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A

20 PARTICULAR PURPOSE. See the GNU General Public License for more details.

22 You should have received a copy of the GNU General Public License

23 along with libDAI in the file COPYING. If not, see http://www.gnu.org/licenses/

25 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

27 Citing libDAI

29 If you write a scientific paper describing research that made substantive use

30 of this program, please:

32 • mention the fashion in which this software was used, including the version

33 number, with a citation to the literature, to allow replication;

34 • mention this software in the Acknowledgements section.

36 An appropriate citation would be:

37 J. M. Mooij (2008) "libDAI 0.2.2: A free/open source C++ library for Discrete

38 Approximate Inference methods", http://www.libdai.org

40 Moreover, as a personal note, I would appreciate it if you would email

41 (citations of) papers referencing this work to joris dot mooij at libdai dot

42 org.

44 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

46 About libDAI

48 libDAI is a free/open source C++ library (licensed under GPL 2+) that provides

49 implementations of various (approximate) inference methods for discrete

50 graphical models. libDAI supports arbitrary factor graphs with discrete

51 variables; this includes discrete Markov Random Fields and Bayesian Networks.

53 The library is targeted at researchers. To be able to use the library, a good

54 understanding of graphical models is needed.

56 The best way to use libDAI is by writing C++ code that invokes the library; in

57 addition, part of the functionality is accessibly by using the

59 • command line interface

60 • (limited) MatLab interface

61 • (experimental) python interface

62 • (experimental) octave interface.

64 libDAI can be used to implement novel (approximate) inference algorithms and to

65 easily compare the accuracy and performance with existing algorithms that have

66 been implemented already.

68 Features

70 Currently, libDAI supports the following (approximate) inference methods:

72 • Exact inference by brute force enumeration;

73 • Exact inference by junction-tree methods;

74 • Mean Field;

75 • Loopy Belief Propagation [KFL01];

76 • Tree Expectation Propagation [MiQ04];

77 • Generalized Belief Propagation [YFW05];

78 • Double-loop GBP [HAK03];

79 • Various variants of Loop Corrected Belief Propagation [MoK07, MoR05];

80 • Gibbs sampler;

81 • Conditioned BP [EaG09].

83 These inference methods can be used to calculate partition sums, marginals over

84 subsets of variables, and MAP states (the joint state of variables that has

85 maximum probability).

87 In addition, libDAI supports parameter learning of conditional probability

88 tables by Expectation Maximization.

90 Limitations

92 libDAI is not intended to be a complete package for approximate inference.

93 Instead, it should be considered as an "inference engine", providing various

94 inference methods. In particular, it contains no GUI, currently only supports

95 its own file format for input and output (although support for standard file

96 formats may be added later), and provides very limited visualization

97 functionalities. The only learning method supported currently is Expectation

98 Maximization (or Maximum Likelihood if no data is missing) for learning factor

99 parameters.

101 Rationale

103 In my opinion, the lack of open source "reference" implementations hampers

104 progress in research on approximate inference. Methods differ widely in terms

105 of quality and performance characteristics, which also depend in different ways

106 on various properties of the graphical models. Finding the best approximate

107 inference method for a particular application therefore often requires

108 empirical comparisons. However, implementing and debugging these methods takes

109 a lot of time which could otherwise be spent on research. I hope that this code

110 will aid researchers to be able to easily compare various (existing as well as

111 new) approximate inference methods, in this way accelerating research and

112 stimulating real-world applications of approximate inference.

114 Language

116 Because libDAI is implemented in C++, it is very fast compared with

117 implementations in MatLab (a factor 1000 faster is not uncommon). libDAI does

118 provide a (limited) MatLab interface for easy integration with MatLab. It also

119 provides a command line interface and experimental python and octave interfaces

120 (thanks to Patrick Pletscher).

122 Compatibility

124 The code has been developed under Debian GNU/Linux with the GCC compiler suite.

125 libDAI compiles successfully with g++ versions 3.4, 4.1, 4.2 and 4.3.

127 libDAI has also been successfully compiled with MS Visual Studio 2008 under

128 Windows (but not all build targets are supported yet) and with Cygwin under

129 Windows.

131 Finally, libDAI has been compiled successfully on MacOS X.

133 Downloading libDAI

135 The libDAI sources and documentation can be downloaded from the libDAI website:

136 http://www.libdai.org.

138 Mailing list

140 The Google group "libDAI" (http://groups.google.com/group/libdai) can be used

141 for getting support and discussing development issues.

143 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

145 Building libDAI under UNIX variants (Linux / Cygwin / Mac OS X)

147 You need:

149 • a recent version of gcc (at least version 3.4)

150 • GNU make

151 • doxygen

152 • graphviz

153 • recent boost C++ libraries (at least version 1.34, or 1.37 for cygwin;

154 version 1.37 shipped with Ubuntu 9.04 is known not to work)

156 On Debian/Ubuntu, you can easily install all these packages with a single

157 command:

159 apt-get install g++ make doxygen graphviz libboost-dev libboost-graph-dev libboost-program-options-dev

161 (root permissions needed).

163 On Mac OS X (10.4 is known to work), these packages can be installed easily via

164 MacPorts. If MacPorts is not already installed, install it according to the

165 instructions at http://www.macports.org/. Then, a simple

167 sudo port install gmake boost doxygen graphviz

169 should be enough to install everything that is needed.

171 On Cygwin, the prebuilt Cygwin package boost-1.33.1-x is known not to work. You

172 can however obtain the latest boost version (you need at least 1.37.0) from

173 http://www.boost.org/ and compile/install it with:

175 ./configure

176 make

177 make install

180 To build the libDAI source, first copy a template Makefile.* to Makefile.conf

181 (for example, copy Makefile.LINUX to Makefile.conf if you use GNU/Linux). Then,

182 edit the Makefile.conf template to adapt it to your local setup. Especially

183 directories may differ from system to system. Finally, run

185 make

187 The build includes a regression test, which may take a while to complete.

189 If the build was successful, you can test the example program:

191 examples/example tests/alarm.fg

193 or the more elaborate test program:

195 tests/testdai --aliases tests/aliases.conf --filename tests/alarm.fg --methods JTREE_HUGIN BP_SEQMAX

197 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

199 Building libDAI under Windows

201 You need:

203 • A recent version of MicroSoft Visual Studio (2008 works)

204 • recent boost C++ libraries (version 1.34 or higher)

205 • GNU make (can be obtained from http://gnuwin32.sourceforge.net)

207 For the regression test, you need:

209 • GNU diff, GNU sed (can be obtained from http://gnuwin32.sourceforge.net)

211 To build the source, copy Makefile.WINDOWS to Makefile.conf. Then, edit

212 Makefile.conf to adapt it to your local setup. Finally, run (from the command

213 line)

215 make

217 The build includes a regression test, which may take a while to complete.

219 If the build was successful, you can test the example program:

221 example tests\alarm.fg

223 or the more elaborate test program:

225 tests\testdai --aliases tests\aliases.conf --filename tests\alarm.fg --methods JTREE_HUGIN BP_SEQMAX

227 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

229 Building the libDAI MatLab interface

231 You need:

233 • MatLab

234 • The platform-dependent requirements described above

236 First, you need to build the libDAI source as described above for your

237 platform. By default, the MatLab interface is disabled, so before compiling the

238 source, you have to enable it in the Makefile.conf by setting

240 WITH_MATLAB=true

242 Also, you have to configure the MatLab-specific parts of Makefile.conf to match

243 your system (in particular, the Makefile variables ME, MATLABDIR and MEX). The

244 MEX file extension depends on your platform; for a 64-bit linux x86_64 system

245 this would be "ME=.mexa64", for a 32-bit linux x86 system "ME=.mexglx". If you

246 are unsure about your MEX file extension: it needs to be the same as what the

247 MatLab command "mexext" returns. The required MEX files are built by issuing

249 make

251 from the command line. The MatLab interface is much less powerful than using

252 libDAI from C++. There are two reasons for this: (i) it is boring to write MEX

253 files; (ii) the large performance penalty paid when large data structures (like

254 factor graphs) have to be converted between their native C++ data structure to

255 something that MatLab understands.

257 A simple example of how to use the MatLab interface is the following (entered

258 at the MatLab prompt), which performs exact inference by the junction tree

259 algorithm and approximate inference by belief propagation on the ALARM network:

261 cd path_to_libdai/matlab

262 [psi] = dai_readfg ('../examples/alarm.fg');

263 [logZ,q,md,qv,qf] = dai (psi, 'JTREE', '[updates=HUGIN,verbose=0]')

264 [logZ,q,md,qv,qf] = dai (psi, 'BP', '[updates=SEQMAX,tol=1e-9,maxiter=10000,logdomain=0]')

266 where "path_to_libdai" has to be replaced with the directory in which libDAI

267 was installed. For other algorithms and some default parameters, see the file

268 tests/aliases.conf.

270 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

272 Building the documentation

274 Install doxygen, graphviz and a TeX distribution and use

276 make doc

278 to build the documentation. If the documentation is not clear enough, feel free

279 to send me an email (or even better, to improve the documentation and send a

280 patch!). The documentation can also be browsed online at http://www.libdai.org.