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

2 ==================================================================================

4 v 0.2.1 - May 26, 2008

7 Copyright (C) 2006-2008 Joris Mooij [j dot mooij at science dot ru dot nl]

8 Radboud University Nijmegen, The Netherlands

11 ----------------------------------------------------------------------------------

12 This file is part of libDAI.

14 libDAI is free software; you can redistribute it and/or modify

15 it under the terms of the GNU General Public License as published by

16 the Free Software Foundation; either version 2 of the License, or

17 (at your option) any later version.

19 libDAI is distributed in the hope that it will be useful,

20 but WITHOUT ANY WARRANTY; without even the implied warranty of

21 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the

22 GNU General Public License for more details.

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

25 along with libDAI; if not, write to the Free Software

26 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA

27 ----------------------------------------------------------------------------------

30 SCIENTISTS: please be aware that the fact that this program is released as Free

31 Software does not excuse you from scientific propriety, which obligates you to

32 give appropriate credit! If you write a scientific paper describing research

33 that made substantive use of this program, it is your moral obligation as a

34 scientist to (a) mention the fashion in which this software was used, including

35 the version number, with a citation to the literature, to allow replication;

36 (b) mention this software in the Acknowledgements section. The appropriate

37 citation is: J. M. Mooij (2008) libDAI: A free/open source C++ library for

38 Discrete Approximate Inference methods, http://mloss.org/software/view/77/.

39 Moreover, as a personal note, I would appreciate it if you would email me with

40 citations of papers referencing this work so I can mention them to my funding

41 agent and tenure committee.

43 What is libDAI?

44 ---------------

45 libDAI is a free/open source C++ library (licensed under GPL, see the file

46 COPYING for more details) that provides implementations of various

47 (deterministic) approximate inference methods for discrete graphical models.

48 libDAI supports arbitrary factor graphs with discrete variables (this includes

49 discrete Markov Random Fields and Bayesian Networks).

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

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

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

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

55 formats may be added), and provides no visualization.

57 Because libDAI is implemented in C++, it is very fast compared with e.g. MatLab

58 implementations. libDAI does provide a MatLab interface for easy integration

59 with MatLab. Currently, libDAI supports the following deterministic approximate

60 inference methods:

62 * Mean Field

63 * (Loopy) Belief Propagation

64 * Tree Expectation Propagation

65 * Generalized Belief Propagation

66 * Double-loop GBP

67 * Loop Corrected Approximate Inference

69 Exact inference by JunctionTree is also provided.

71 Many of these algorithms are not yet available in similar open source software,

72 to the best of the author's knowledge (open source packages supporting both

73 directed and undirected graphical models are Murphy's BNT, Intel's PNL and gR).

75 The library is targeted at researchers; to be able to use the library, a good

76 understanding of graphical models is needed. However, the code will hopefully

77 find its way into real-world applications as well.

80 Rationale

81 ---------

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

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

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

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

86 inference method for a particular application therefore often requires

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

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

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

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

91 stimulating real-world applications of approximate inference.

94 Releases

95 --------

96 Releases can be obtained from http://www.mbfys.ru.nl/~jorism/libDAI.

97 License: GNU Public License v2 (or higher).

99 libDAI-0.2 December 1, 2006

100 libDAI-0.2.1 May 26, 2008

103 Acknowledgments

104 ---------------

105 The development reported here is part of the Interactive Collaborative

106 Information Systems (ICIS) project, supported by the Dutch Ministry of Economic

107 Affairs, grant BSIK03024. I would like to thank Martijn Leisink for providing

108 the basis on which libDAI has been built.

111 Known issues

112 ------------

113 Due to a bug in GCC 3.3.x and earlier (http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20358)

114 it doesn't compile with these versions (it does compile with GCC version 3.4 and higher).

115 Workaround: replace the two NAN's in factor.h causing the error messages by -1.

118 Documentation

119 -------------

120 Almost nonexistant. But I'm working on it. In the meantime, I'll provide limited support

121 by email. The following gives an overview of different methods and their properties

122 (can be slightly obsolete):

124 BP

125 updates UpdateType SEQFIX,SEQRND,SEQMAX,PARALL

126 tol double

127 maxiter size_t

128 verbose size_t

129 MF

130 tol double

131 maxiter size_t

132 verbose size_t

133 HAK

134 clusters MIN,DELTA,LOOP

135 loopdepth

136 doubleloop bool

137 tol double

138 maxiter size_t

139 verbose size_t

140 JTREE

141 updates UpdateType HUGIN,SHSH

142 verbose size_t

143 MR

144 updates UpdateType FULL,LINEAR

145 inits InitType RESPPROP,CLAMPING,EXACT

146 verbose size_t

147 TREEEP

148 type TypeType ORG,ALT

149 tol double

150 maxiter size_t

151 verbose size_t

152 LC

153 cavity CavityType FULL,PAIR,PAIR2,UNIFORM

154 updates UpdateType SEQFIX,SEQRND(,NONE)

155 reinit bool

156 cavainame string

157 cavaiopts Properties

158 tol double

159 maxiter size_t

160 verbose size_t

164 Quick start

165 -----------

166 You need:

167 - a recent version of gcc (version 3.4 at least)

168 - the Boost C++ libraries (under Debian/Ubuntu you can install them using

169 "apt-get install libboost-dev libboost-program-options-dev")

170 - GNU make

172 To build the source, edit the Makefile and then run

174 make

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

178 ./example tests/alarm.fg

180 or the more elaborate test program:

182 tests/test --aliases tests/aliases.conf --filename tests/alarm.fg --methods JTREE_HUGIN BP_SEQMAX

184 A description of the factor graph (.fg) file format can be found in the file FILEFORMAT.