1 libDAI - A free/open source C++ library for Discrete Approximate Inference
3 -------------------------------------------------------------------------------
6 Date: July 17, 2015 - or later
7 See also: http://www.libdai.org
9 -------------------------------------------------------------------------------
13 libDAI is free software; you can redistribute it and/or modify it under the
14 terms of the BSD 2-clause license (also known as the FreeBSD license), which
15 can be found in the accompanying LICENSE file.
17 [Note: up to and including version 0.2.7, libDAI was licensed under the GNU
18 General Public License (GPL) version 2 or higher.]
21 -------------------------------------------------------------------------------
25 If you write a scientific paper describing research that made substantive use
26 of this library, please cite the following paper describing libDAI:
28 libDAI: A free & open source C++ library for Discrete Approximate Inference in
30 Journal of Machine Learning Research, 11(Aug):2169-2173, 2010.
31 In BiBTeX format (for your convenience):
33 @article{Mooij_libDAI_10,
34 author = {Joris M. Mooij},
35 title = {lib{DAI}: A Free and Open Source {C++} Library for Discrete Approximate Inference in Graphical Models},
36 journal = {Journal of Machine Learning Research},
41 url = "http://www.jmlr.org/papers/volume11/mooij10a/mooij10a.pdf"
44 Moreover, as a personal note, I would appreciate it to be informed about any
45 publications using libDAI at joris dot mooij at libdai dot org.
47 -------------------------------------------------------------------------------
51 libDAI is a free/open source C++ library that provides implementations of
52 various (approximate) inference methods for discrete graphical models. libDAI
53 supports arbitrary factor graphs with discrete variables; this includes
54 discrete Markov Random Fields and Bayesian Networks.
56 The library is targeted at researchers. To be able to use the library, a good
57 understanding of graphical models is needed.
59 The best way to use libDAI is by writing C++ code that invokes the library; in
60 addition, part of the functionality is accessibly by using the
62 * command line interface
63 * (limited) MatLab interface
64 * (experimental) python interface
65 * (experimental) octave interface.
67 libDAI can be used to implement novel (approximate) inference algorithms and to
68 easily compare the accuracy and performance with existing algorithms that have
69 been implemented already.
71 A solver using libDAI was amongst the three winners of the UAI 2010 Approximate
72 Inference Challenge (see http://www.cs.huji.ac.il/project/UAI10/ for more
73 information). The full source code is provided as part of the library.
77 Currently, libDAI supports the following (approximate) inference methods:
79 * Exact inference by brute force enumeration;
80 * Exact inference by junction-tree methods;
82 * Loopy Belief Propagation [KFL01];
83 * Fractional Belief Propagation [WiH03];
84 * Tree-Reweighted Belief Propagation [WJW03];
85 * Tree Expectation Propagation [MiQ04];
86 * Generalized Belief Propagation [YFW05];
87 * Double-loop GBP [HAK03];
88 * Various variants of Loop Corrected Belief Propagation [MoK07, MoR05];
90 * Conditioned Belief Propagation [EaG09];
91 * Decimation algorithm.
93 These inference methods can be used to calculate partition sums, marginals over
94 subsets of variables, and MAP states (the joint state of variables that has
97 In addition, libDAI supports parameter learning of conditional probability
98 tables by Expectation Maximization.
102 libDAI is not intended to be a complete package for approximate inference.
103 Instead, it should be considered as an "inference engine", providing various
104 inference methods. In particular, it contains no GUI, currently only supports
105 its own file format for input and output (although support for standard file
106 formats may be added later), and provides very limited visualization
107 functionalities. The only learning method supported currently is Expectation
108 Maximization (or Maximum Likelihood if no data is missing) for learning factor
113 In my opinion, the lack of open source "reference" implementations hampers
114 progress in research on approximate inference. Methods differ widely in terms
115 of quality and performance characteristics, which also depend in different ways
116 on various properties of the graphical models. Finding the best approximate
117 inference method for a particular application therefore often requires
118 empirical comparisons. However, implementing and debugging these methods takes
119 a lot of time which could otherwise be spent on research. I hope that this code
120 will aid researchers to be able to easily compare various (existing as well as
121 new) approximate inference methods, in this way accelerating research and
122 stimulating real-world applications of approximate inference.
126 Because libDAI is implemented in C++, it is very fast compared with
127 implementations in MatLab (a factor 1000 faster is not uncommon). libDAI does
128 provide a (limited) MatLab interface for easy integration with MatLab. It also
129 provides a command line interface and experimental python and octave interfaces
130 (thanks to Patrick Pletscher).
134 The code has been developed under Debian GNU/Linux with the GCC compiler suite.
135 libDAI compiles successfully with g++ versions 3.4 up to 4.7 (both 32 and 64
138 libDAI has also been successfully compiled with MS Visual Studio 2008 under
139 Windows, MS Visual Studio 2010 under Windows 64, and with Cygwin under Windows.
141 Finally, libDAI has been compiled successfully on MacOS X (both 32 and 64
146 The libDAI sources and documentation can be downloaded from the libDAI website:
147 http://www.libdai.org.
151 The Google group "libDAI" (http://groups.google.com/group/libdai) can be used
152 for getting support and discussing development issues.
154 -------------------------------------------------------------------------------
156 Building libDAI under UNIX variants (Linux / Cygwin / Mac OS X)
162 * a recent version of gcc (at least version 3.4)
164 * recent boost C++ libraries (at least version 1.37; however, version 1.37
165 shipped with Ubuntu 9.04 is known not to work)
166 * GMP library (or the Windows port called MPIR, for 64 bits builds MPIR 2.5.0
168 * doxygen (only for building the documentation)
169 * graphviz (only for using some of the libDAI command line utilities)
170 * CImg library (only for building the image segmentation example)
172 On Debian/Ubuntu, you can easily install the required packages with a single
175 apt-get install g++ make doxygen graphviz libboost-dev libboost-graph-dev libboost-program-options-dev libboost-test-dev libgmp-dev cimg-dev
177 (root permissions needed).
179 On Mac OS X (10.4 is known to work), these packages can be installed easily via
180 MacPorts. If MacPorts is not already installed, install it according to the
181 instructions at http://www.macports.org/. Then, a simple
183 sudo port install gmake boost gmp doxygen graphviz
185 should be enough to install everything that is needed.
187 On Cygwin, the prebuilt Cygwin package boost-1.33.1-x is known not to work. You
188 can however obtain the latest boost version (you need at least 1.37.0) from
189 http://www.boost.org/ and build it as described in the next subsection.
191 Building boost under Cygwin
193 * Download the latest boost libraries from http://www.boost.org
194 * Build the required boost libraries using:
196 ./bootstrap.sh --with-libraries=program_options,math,graph,test --prefix=/boost_root/
199 * In order to use dynamic linking, the boost .dll's should be somewhere in
200 the path. This can be achieved by a command like:
202 export PATH=$PATH:/boost_root/stage/lib
206 To build the libDAI source, first copy a template Makefile.* to Makefile.conf
207 (for example, copy Makefile.LINUX to Makefile.conf if you use GNU/Linux). Then,
208 edit the Makefile.conf template to adapt it to your local setup. In case you
209 want to use Boost libraries which are installed in non-standard locations, you
210 have to tell the compiler and linker about their locations (using the -I, -L
211 flags for GCC; also you may need to set the LD_LIBRARY_PATH environment
212 variable correctly before running libDAI binaries). Platform independent build
213 options can be set in Makefile.ALL. Finally, run
217 The build includes a regression test, which may take a while to complete.
219 If the build is successful, you can test the example program:
221 examples/example tests/alarm.fg
223 or the more extensive test program:
225 tests/testdai --aliases tests/aliases.conf --filename tests/alarm.fg --methods JTREE_HUGIN BP_SEQMAX
227 -------------------------------------------------------------------------------
229 Building libDAI under Windows
235 * A recent version of MicroSoft Visual Studio (2008 is known to work)
236 * recent boost C++ libraries (version 1.37 or higher)
237 * GMP or MPIR library (for 64-bits builds, MPIR 2.5.0 or higher is needed)
238 * GNU make (can be obtained from http://gnuwin32.sourceforge.net)
239 * CImg library (only for building the image segmentation example)
241 For the regression test, you need:
243 * GNU diff, GNU sed (can be obtained from http://gnuwin32.sourceforge.net)
245 Building boost under Windows
247 Because building boost under Windows is tricky, I provide some guidance here.
249 * Download the boost zip file from http://www.boost.org/users/download and
251 * Download the bjam executable from http://www.boost.org/users/download and
252 unpack it somewhere else.
253 * Download Boost.Build (v2) from http://www.boost.org/docs/tools/build/
254 index.html and unpack it yet somewhere else.
255 * Edit the file boost-build.jam in the main boost directory to change the
256 BOOST_BUILD directory to the place where you put Boost.Build (use UNIX /
257 instead of Windows \ in pathnames).
258 * Copy the bjam.exe executable into the main boost directory. Now if you
259 issue "bjam --version" you should get a version and no errors. Issueing
260 "bjam --show-libraries" will show the libraries that will be built.
261 * The following command builds the boost libraries that are relevant for
264 bjam --with-graph --with-math --with-program_options --with-test link=static runtime-link=shared
266 Building GMP or MPIR under Windows
268 Information about how to build GPR or MPIR under Windows can be found on the
269 internet. The user has to update Makefile.WINDOWS in order to link with the GPR
270 /MPIR libraries. Note that for 64-bit builds, MPIR 2.5.0 or higher is needed.
274 To build the source, copy Makefile.WINDOWS to Makefile.conf. Then, edit
275 Makefile.conf to adapt it to your local setup. Platform independent build
276 options can be set in Makefile.ALL. Finally, run (from the command line)
280 The build includes a regression test, which may take a while to complete.
282 If the build is successful, you can test the example program:
284 examples\example tests\alarm.fg
286 or the more extensive test program:
288 tests\testdai --aliases tests\aliases.conf --filename tests\alarm.fg --methods JTREE_HUGIN BP_SEQMAX
290 -------------------------------------------------------------------------------
292 Building the libDAI MatLab interface
297 * The platform-dependent requirements described above
299 First, you need to build the libDAI source as described above for your
300 platform. By default, the MatLab interface is disabled, so before compiling the
301 source, you have to enable it in Makefile.ALL by setting
305 Also, you have to configure the MatLab-specific parts of Makefile.conf to match
306 your system (in particular, the Makefile variables ME, MATLABDIR and MEX). The
307 MEX file extension depends on your platform; for a 64-bit linux x86_64 system
308 this would be "ME=.mexa64", for a 32-bit linux x86 system "ME=.mexglx". If you
309 are unsure about your MEX file extension: it needs to be the same as what the
310 MatLab command "mexext" returns. The required MEX files are built by issuing
314 from the command line. The MatLab interface is much less powerful than using
315 libDAI from C++. There are two reasons for this: (i) it is boring to write MEX
316 files; (ii) the large performance penalty paid when large data structures (like
317 factor graphs) have to be converted between their native C++ data structure to
318 something that MatLab understands.
320 A simple example of how to use the MatLab interface is the following (entered
321 at the MatLab prompt), which performs exact inference by the junction tree
322 algorithm and approximate inference by belief propagation on the ALARM network:
324 cd path_to_libdai/matlab
325 [psi] = dai_readfg ('../tests/alarm.fg');
326 [logZ,q,md,qv,qf] = dai (psi, 'JTREE', '[updates=HUGIN,verbose=0]')
327 [logZ,q,md,qv,qf] = dai (psi, 'BP', '[updates=SEQMAX,tol=1e-9,maxiter=10000,logdomain=0]')
329 where "path_to_libdai" has to be replaced with the directory in which libDAI
330 was installed. For other algorithms and some default parameters, see the file
333 -------------------------------------------------------------------------------
335 Building the documentation
337 Install doxygen, graphviz and a TeX distribution and use
341 to build the documentation. If the documentation is not clear enough, feel free
342 to send me an email (or even better, to improve the documentation and send a
343 patch!). The documentation can also be browsed online at http://www.libdai.org.