1 * Added init parameter to HAK/GBP to allow for random initialization

2 * Replaced the standard assert() macro by DAI_ASSERT, which throws a

3 dai::Exception and is even active if NDEBUG is defined

4 * Added a constructor TProb<T>::TProb<T>( const std::vector<S> &v )

5 * Introduced CCNODEBUGFLAGS in Makefile

6 * Updated copyright headers

7 * Added max-product functionality to JTree

8 * [Charlie Vaske] EM bugfix (was using abs() instead of fabs() in

9 determining EM termination, which caused a loss of precision).

10 * [Charlie Vaske] New constructor in Permute for canonical variable ordering

11 * [Charlie Vaske] New constructor in Factor that reorders variables to canonical ordering

12 * [Charlie Vaske] New accessor in Properties for getting all keys

13 * [Charlie Vaske] New function for getting inferred parameters

14 * Cleaned up error handling by introducing the DAI_THROWE macro

15 * Introduced DAI_DEBASSERT macro to abbreviate a common pattern

16 * Changed FactorGraph::clamp and DAIAlg::clamp interfaces (the variable to be

17 clamped is now indicated by its index, not as a Var) and marked the old

18 interface as obsolete

19 * [Patrick Pletscher] Fixed performance issue in FactorGraph::clamp

20 * [Sebastian Nowozin] MEX file dai now also optionally returns the MAP state

21 (only for BP)

22 * [Sebastian Nowozin] Fixed memory leak in MEX file dai

24 git ae0fc30e10be6683cbfc209dcee56f34234a6cb8

25 --------------------------------------------

26 * Added work-around for bug in Boost Graph Library

27 * Improvements of TFactor<T>:

28 - Extended functionality of TFactor<T>::operators +,-,+=,-= to handle different VarSets

29 - Added TFactor<T>::maxMarginal (similar to marginal but with max instead of sum)

30 * Added BipartiteGraph::eraseEdge

31 * Removed all the virtual default constructors *::create(), as they are not used.

32 * Fixed bug in MaxSpanningTree (it wrongly assumed that the tree was not empty).

33 * [Charlie Vaske] Added Expectation Maximization code.

34 * Added MatLab QuickStart to README.

35 * MEX file dai now also returns variable and factor beliefs.

36 * Cleanups and extra documentation of the code contributed by Frederik Eaton

37 * Removed erroneous 'inline' directives in gibbs.cpp

38 * [Frederik Eaton] Added script for automatically generating properties code (used by CBP and BBP)

39 * [Frederik Eaton] Updated doxygen.conf to version 1.5.9

40 * [Frederik Eaton] Added newInfAlgFromString to alldai.h/cpp

41 * [Frederik Eaton] Added FactorGraph::clampVar and FactorGraph::clampFactor

42 * [Frederik Eaton] Changed BP to be able to record order in which messages are sent (for use by BBP)

43 * [Frederik Eaton] Improved documentation of BipartiteGraph::Neighbor

44 * [Frederik Eaton]: Extended InfAlg interface with beliefV() and beliefF() methods;

45 * [Frederik Eaton]: Improved PropertySet class:

46 - Added default constructor

47 - Allow construction from string

48 - Added method allKeys() to produce list of keys

49 - In getStringAs(), check for typeid(ValueType) before typeid(std::string)

50 (this fixes a strange bug for empty string property)

51 * [Frederik Eaton]: Added some utility macros (DAI_PV, DAI_DMSG, DAI_ACCMUT, DAI_IFVERB)

52 and functions (concat(),rnd()) to include/dai/util.h

53 * [Frederik Eaton] Added backwards compatibility layer for edge interface to

54 BipartiteGraph and FactorGraph (which will be obsoleted soon)

55 * [Frederik Eaton] Added BP_dual, BBP and CBP algorithms

56 * [Frederik Eaton] Added Gibbs::state() accessors/mutators

57 * [Frederik Eaton] Fixed Gibbs::getVarDist(size_t) to return uniform

58 distribution if no state is allowed

59 * [Frederik Eaton] Improved parsing code in tests/testdai to allow recursive

60 alias substitutions

61 * Interface changes:

62 TProb<T>::minVal() -> TProb<T>::min()

63 TFactor<T>::minVal() -> TFactor<T>::min()

64 TProb<T>::maxVal() -> TProb<T>::max()

65 TFactor<T>::maxVal() -> TFactor<T>::max()

66 TProb<T>::totalSum() -> TProb<T>sum()

67 TFactor<T>::totalSum() -> TFactor<T>::sum()

68 * [Frederik Eaton] Added methods setUniform (), sumAbs(), argmax() to TProb<T>

69 * [Frederik Eaton] Added TAGS, lib targets to Makefile

70 * [Frederik Eaton] Add useful aliases to aliases.conf

71 * [Frederik Eaton] Removed unnecessary copy constructors and assignment

72 operators

73 * [Frederik Eaton] Change cout to cerr in warnings and error messages

74 * [Giuseppe Passino] Optimised maximum-residual BP by using a reversed ordered

75 set instead of the linear search (which can yield enormous speedups - a

76 factor 500 has been measured on a binary Ising grid with 128x128 variables!)

77 * Added debug assertions to Var which check for inconsistent dimensions of

78 variables with the same labels

79 * [Giuseppe Passino] Added prefix ++ operator to State (State::operator++())

80 * [Giuseppe Passino] Added iterators to FactorGraph (FactorGraph::begin,

81 FactorGraph::end)

82 * [Giuseppe Passino] Added iterators to TFactor (TFactor::begin, TFactor::end)

83 * [Giuseppe Passino] Added iterators to TProb (TProb::begin, TProb::end)

84 * [Giuseppe Passino] Added BP::findMaximum(), which can be used after running

85 max-product BP to construct a global state with maximum probability

86 * Improved Makefile (libDAI now also builds out-of-the-box on MacOSX,

87 thanks to Dan Preston; merged Makefile.win and Makefile.shared into Makefile)

88 * Fixed bug in calcMarginal, calcPairBeliefs, calcPairBeliefsNew where

89 states with zero probability mass were clamped, leading to NaNs or assertion

90 errors (thanks to Dan Preston for reporting this)

91 * toc() now returns POSIX system time with maximum accuracy of microseconds

92 * Exception objects now remember their error code

93 * Added examples/example_sprinkler.cpp

94 * [Peter Gober] Made libDAI build out-of-the-box on Cygwin.

95 * [Frederik Eaton] Added Gibbs sampler to algorithms

96 * Improved documentation of include/dai/{bipgraph.h, prob.h, smallset.h,

97 var.h, varset.h, factor.h, enum.h} and of examples/example.cpp

98 Merged TODO and FILEFORMAT into doxygen documentation

99 * examples/

100 - Moved example.cpp to examples/

101 - Added examples/example_bipgraph.cpp

102 - Added examples/example_varset.cpp

103 * Interface changes:

104 - Replaced TProb<T>::log0() by TProb<T>::log(true)

105 - Replaced TProb<T>::takeLog0() by TProb<T>::takeLog(true)

106 - Replaced TFactor<T>::log0() by TFactor<T>::log(true)

107 - Removed TProb<T>::hasNonPositives()

108 - Renamed smallSet<T> to SmallSet<T>

109 - Removed TFactor<T>::divided_by, use operator/ instead

110 - Removed TFactor<T>::divide, use operator/= instead

111 - Removed TFactor<T>::partSum(const VarSet&), use marginal(const VarSet&,true) instead

112 - Improved constructors of TProb and TFactor to use iterators instead of pointers

113 * Miscellaneous small changes and cleanups:

114 - Changed regression test so that it also works under Windows

115 - Changed output stream operator<< for Var and VarSet

116 - Added TProb::draw() function, which draws a random index

117 - Cleanup of matlab interface code

118 - Small improvement of utils/fginfo

119 - Small cleanup of BP code

120 - Switched Makefile.win to GNU Make syntax

123 libDAI-0.2.2 (2008-09-30)

124 -------------------------

126 New features:

127 * Approximate inference methods now report the number of iterations needed.

128 * Added damping to various algorithms to improve convergence properties.

129 * Added more features to utils/createfg for creating factor graphs.

130 * Added ExactInf class for brute force exact inference.

131 * [Giuseppe Pasino] Added "logdomain" property to BP, a boolean that controls

132 whether calculations are done in the log-domain or in the linear domain;

133 doing calculations in the log-domain may help if the numerical range

134 of a double is too small.

135 * [Claudio Lima] Added Max-Product functionality to BP.

136 * Improved documentation.

138 Improved architecture:

139 * Added Exceptions framework.

140 * Pervasive change of BipartiteGraph implementation (based on an idea by

141 Giuseppe Passino). BipartiteGraph no longer stores the node properties

142 (former _V1 and _V2), nor does it store a dense adjacency matrix anymore,

143 nor an edge list. Instead, it stores the graph structure as lists of

144 neighboring nodes. This yields a significant memory/speed improvement for

145 large factor graphs, and is more elegant as well. Iterating over neighbors is

146 made easy by using boost::foreach.

147 * Added conditional compilation of inference methods.

148 * VarSet is now implemented using a std::vector<Var> instead of a

149 std::set<Var>, which yields a significant speed improvement. Furthermore,

150 the implementation has been generalized, resulting in the small_set<T> class

151 which can be used to represent sets of small cardinality; VarSet is the

152 specialization with T = Var.

153 * Improved ClusterGraph implementation, yielding significant speedups

154 for the JunctionTree algorithm on large factorgraphs.

156 Code cleanup:

157 * Moved everything into namespace "dai".

158 * Renamed DEBUG to DAI_DEBUG to avoid conflicts.

159 * Replaced ENUM2,ENUM3,ENUM4,ENUM5,ENUM6 by single DAI_ENUM macro.

160 * Removed utils/remove_short_loops and matlab/remove_short_loops.

161 * Replaced sub_nb class in mr.h by boost::dynamic_bitset.

162 * Improved index.h:

163 - Renamed Index -> IndexFor

164 - Added some .reserve()'s to IndexFor methods which yields a

165 25% speedup of testregression

166 - Replaced multind by Permute

167 - Added MultiFor

168 - Added State

169 * New funcstionality of factor.h.

170 * Moved Properties and MaxDiff frameworks from InfAlg to each individual

171 inference algorithm, because the Properties framework was not as

172 convenient as I hoped, and not every inference algorithm needs a maxdiff

173 variable. Also, replaced some FactorGraph functionality in InfAlg by a

174 function that returns the FactorGraph. The result is cleaner (less

175 entangled) code.

176 * Removed x2x.

177 * Replaced Complex with real numbers (negative potentials are just too rare

178 to warrant the additional "complexity" :)).

180 Miscellaneous improvements:

181 * Now compiles also with MS Visual C++ (thanks to Jiuxiang Hu) and with

182 GCC under cygwin.

183 * Contributions by Peter Gober:

184 - Renamed variable _N in mr.* for compatibility with g++ under cygwin.

185 * Misc contributions by Giuseppe Passino:

186 - removed "using namespace std;" from header files - bad practice;

187 - moved header files in include/dai and sources in src;

188 - changed #ifndefs to GNU style;

189 - added extra warning checks (-W -Wextra) and fixed resulting warnings;

190 - dai::TProb:

191 o removed copy constructor and assignment operators (redundant);

192 o implementation of some methods via STL algorithms;

193 o added methods takeExp, takeLog, takeLog0 for transformation in-place;

194 o explicit constructor (prevents implicit conversion from size_t to TProb);

195 o added operator+,+=,-,-=, with argument T (for calculations in log-scale);

196 * Misc contributions by Christian Wojek:

197 - New FactorGraph constructor that constructs from given ranges of factors

198 and variables;

199 - Optimization of FactorGraph constructors using tr1::unordered_map.

200 * FactorGraph constructors no longer check for short loops (huge speed

201 increase for large factor graphs), nor for negative entries. Also, the

202 normtype is now Prob::NORMPROB by default.

203 * Improved MaxSpanningTreePrims algorithm (uses boost::graph).

205 Interface changes:

206 * VarSet::

207 - VarSet::stateSpace() -> nrStates(const VarSet &)

208 - VarSet( const std::set<Var> ) -> VarSet( begin, end, sizeHint=0 )

209 - VarSet( const std::vector<Var> ) -> VarSet( begin, end, sizeHint=0 )

210 - removed bool operator||

211 - operator&&(const VarSet&) -> intersects(const VarSet&)

212 - operator&&(const Var&) -> contains(const Var&)

213 * FactorGraph::

214 - delta(const Var &) -> delta(size_t)

215 - Delta(const Var &) -> Delta(size_t)

216 - makeCavity(const Var &) -> makeCavity(size_t)

217 - vars() -> vars

218 - factors() -> factors

219 - removed MakeFactorCavity(size_t)

220 - removed ExactMarginal(const VarSet &)

221 - removed ExactlogZ()

222 - removed updatedFactor(size_t)

223 - removed _normtype and NormType()

224 - removed hasShortLoops(...) and removeShortLoops(...)

225 - WriteToDotFile(const char *filename) -> printDot( std::ostream& os )

226 - undoProb(size_t) -> restoreFactor(size_t)

227 - saveProb(size_t) -> backupFactor(size_t)

228 - undoProbs(const VarSet &) -> restoreFactors(const VarSet &)

229 - saveProbs(const VarSet &) -> backupFactors(const VarSet &)

230 - ReadFromFile(const char*) returns void (throws on error)

231 - WriteToFile(const char*) returns void (throws on error)

232 - removed hasNegatives()

233 * RegionGraph::

234 - nr_ORs() -> nrORs()

235 - nr_IRs() -> nrIRs()

236 - ORs() -> ORs

237 - IRs() -> IRs

238 * *::Regenerate() -> *::construct()

239 * Renamed Index -> IndexFor

240 * Diffs:

241 - max() -> maxDiff()

242 - max_size() -> maxSize()

243 * Prob::max() -> Prob::maxVal()

244 * Factor::

245 - max() -> maxVal()

246 - part_sum() -> partSum()

247 * toc() in util.h now returns seconds as a double

248 * VarSet::operator&&

249 * Properties -> PropertySet

252 libDAI-0.2.1 (2008-05-26)

253 -------------------------

255 Bugfix release.

256 * added missing cstdio header in util.h

257 * fixed Properties in MR_CLAMPING_* and MR_EXACT_*

258 * added description of the factor graph fileformat

259 * improved Makefile

262 libDAI-0.2.0 (2006-11-30)

263 -------------------------

265 First public release.

268 0.1.5 (2006-11-30)

269 ------------------

271 Regressions

273 - tests/testlcbp and tests/testlcbp are broken.

274 - EXACT method does not work anymore.

275 - The Properties framework gives a speed penalty because of the lookup

276 costs involved; inner loops should be optimized.

278 General framework

280 - DAIAlg is now a template class; typedefs for DAIAlg<FactorGraph> and for

281 DAIAlg<RegionGraph> are provided. In this way, we do not have to write "wrapper"

282 functions to forward functionality from either FactorGraph or RegionGraph

283 to DAIAlg. Functionality like clamping can be implemented in FactorGraph

284 and in RegionGraph and no explicit interface is needed in descendants.

285 - New abstract base class InfAlg added, representing an inference algorithm,

286 from which DAIAlg<T> inherits. This solves the incompatibility problems of

287 DAIAlg<T> for different T (e.g. DAIAlg<FactorGraph> was incompatible with

288 DAIAlg<RegionGraph>). More work is required to reduce code duplication

289 (make FactorGraph part of InfAlg).

290 - Added generic interface (nrVars(), Vars(), nrFactors(), factor(size_t),

291 beliefs(), belief(VarSet &), ...) to InfAlg and descendants.

292 - Added a saveProbs/undoProbs interface to InfAlg and descendants that enables

293 one to save a few factors, modify them (e.g. clamp them), and then restore them

294 to their old values. Undo should also init the corresponding messages / beliefs.

295 This can be used if a given factor graph repeatedly needs to be clamped in

296 different ways and an approximation method is run for each clamping; using the

297 saveProbs/undoProbs can give a significant speed increase.

298 - Switched to a general Properties framework that handles the parameters of

299 all inference methods in a uniform manner. The Properties class is a map of

300 several properties in boost::any objects, indexed by their names (strings).

301 It can read from a stream and write to a stream. It is recursive, in the sense

302 that a Properties object can hold a variable of type Properties as well.

303 - Added a generic way of constructing inference algorithms from a factor graph,

304 name and properties object. Added the newInfAlg function which constructs

305 the requested object. This is used by LCBP, the Matlab interface and the

306 command line (test) interface.

307 - Added a generic enum framework for enum parameters. Although implemented as a

308 hack, it is quite useful in that it drastically simplifies and reduces the

309 amount of code for handling enum parameters.

310 - Provided generic functions for calculating marginals in different ways that

311 work for all approximate inference methods.

313 Bugfixes

315 - Fixed GBP free energy.

316 - Fixed bug in junctiontree (it didn't init the _vars variable).

317 - Corrected two bugs in operator&& and operator|| in VarSet (they returned

318 the logical NOT of what they should return).

319 - Fixed bug in RegionGraph::RecomputeOR(s).

320 - Fixed bug in utils/create_dreg_fg:

321 graph structure was not random for given parameters (forgot to call srand()).

322 - TreeEP bug workaround: use the complete junction tree instead of a subtree.

323 - Fixed bug in JTree::HUGIN() and JTree:ShaferShenoy() in case of junction tree

324 that consists of one outer region only.

325 - Fixed INIT bug in LCBP2::UpdatePancake().

326 - Fixed MaxDiffs flow (except for MR).

328 New functionality

330 - HAK supports several default cluster choices:

331 minimal (only factors)

332 delta (Markov blankets)

333 loop (all loops consisting of loops consisting of <loopdepth> or less variables)

334 Only the maximal clusters are used as outer clusters.

335 - Implemented TreeEP. It generalizes the heuristic method described in the

336 Minka & Qi paper for obtaining a tree with the most relevant interactions to

337 higher order interactions. Almost all optimizations described in the Minka & Qi

338 paper are used, except that evidence is passed over the whole tree instead of

339 relevant subsets (the latter is almost implemented but buggy). Also added

340 alternative (worst-case) algorithm that uses a maximum spanning tree on the

341 weighted graph where the weight between neighbours i and j is given by

342 N(psi,i,j), where psi is the product of all the factors involving both i and j

343 (which is an upper bound on the effective interaction between i and j).

344 - Implemented MR (MontanariRizzo) based on Bastian's code, but extended it

345 to be able to handle connectivities larger than 3 (in principle, up to 32).

346 It supports different initialization methods (the original RESPPROP,

347 the CLAMPING method and EXACT which uses JTree) and different update methods

348 (FULL and LINEAR).

349 - Implemented LCBP2, an analogon of LCBP which represents pancakes as little

350 networks and uses some approximate inference method on them for calculating

351 marginals.

352 - Now there are several LCBP variants (LCBP, LCBPI, LCBPJ, LCBPK, LCBPL);

353 LCBPJ works only for pairwise, LCBPK is LCBP improved for higher order

354 interactions and LCBPL is LCBPI improved for higher-order interactions.

355 - Wrote one single program utils/createfg for creating various types of

356 random factor graphs.

357 - Wrote utility to visualize factor graphs using graphviz.

358 (it uses the BOOST Program Options library)

359 - Added fginfo utility that displays some info about a .fg file.

360 - Implemented Factor::strength function that calculates the potential strength

361 N(psi,i,j) between variables i and j as described in cs.IT:0504030

362 - Wrote a general MatLab interface matlab/dai (similar to tests/test);

363 this unified the matlab functions dai, dai_bp, dai_mf, dai_jt, dai_tep, dai_cvm.

364 - Added MATLAB routine that returns contraction matrix A for BP convergence analysis.

365 - Implemented a MATLAB interface ai_potstrength for Factor::strength

366 - Added Martijn's x2x

368 Improvements of existing code

370 - Reimplemented RegionGraph and descendants: a RegionGraph ISA FactorGraph

371 and also a BipartiteGraph<FRegion,Region>. It now also keeps a map that

372 associates outer region indices to factor indices (no powers yet, this

373 is deemed superfluous) and provides functions to recompute (some of) the

374 outer regions from the factors.

375 - InfAlg descendants run() methods now stop immediately and return NAN in case

376 they detect NANs. Only BP does not do NAN checking for performance reasons.

377 - LCBP now works with factors containing zeroes (by defining x/0 = 0).

378 - HAK, GBP and DoubleLoop now conform to the standards for verbose reporting,

379 timing and convergence criteria.

380 - Implemented logZ() for JTree. It does the calculation during message-passing.

381 - Marginal2ndO now optionally divides by the single node beliefs (to the power n-2);

382 hopefully this will give more accurate approximations.

383 - Marginal and Marginal2ndO (optionally) use the new saveProbs/undoProbs functionality

384 for a faster way of calculating marginals, which does not require a call to init()

385 nor cloning the whole object for each clamping. This leads to a significant speedup.

386 - LCBP (and LCBP2) now have complete flexibility in the specification of the

387 inner method, i.e. the method used to generate the initial cavity approximations.

388 One can pass two strings, a name and a properties string, and LCBP simply invokes

389 newInfAlg to construct the corresponding inference algorithm and uses the generic

390 marginal functions to approximate cavity marginals.

391 - Replaced the global "method" variable by local properties and removed ai.h

392 - Added some methods to Factor (operators *, *=, /, /= with doubles as

393 second argument, operators -, +=, -= with other Factors as second

394 arguments, randomize(), RemoveFirstOrderInteractions) and similar

395 operations to Prob

396 - Moving towards boost::program_options for handling command line arguments

397 (tests/test is done).

398 - Renamed some FactorGraph methods:

399 nr_vars -> nrVars

400 nr_factors -> nrFactors

401 varind -> findVar

402 factorind -> findFactor

403 makeFacCavity -> makeFactorCavity

404 - LCBP_SEQMAXRES has been removed because it did strange things.

405 - Implemented RandomDRegularGraph

406 - Implemented JTree::calcMarginal for marginals that are not confined

407 within one cluster (using cut-set conditioning).

408 - Added isConnected() method to FactorGraph (some methods do not work with

409 disconnected factor graphs).

410 - Pair beliefs are now calculated in a symmetrical way by calcPairBeliefs

411 - Removed single node interaction "correction" code from clamping methods

412 - Removed calcCavityDist and calcCavityDist2ndO

413 - No longer depends on GSL.

414 - Increased portability by combining platform dependant utility functions

415 in util.{h,cpp}.

416 - Wrote *.m files providing help

418 Testing framework

420 - Made a new and significantly improved testing framework that provides most

421 functionality from the command line.

422 - The basis is provided by tests/test, which uses the newInfAlg functionality

423 and enables the user to easily compare from the command line different

424 inference methods on a given factorgraph. All parameters can be specified.

425 Output consists of CPU time, average and maximum single variable marginal

426 errors, relative logZ error and MaxDiff().

427 - tests/aliases.conf contains aliases for standard combinations of methods

428 and their options (which can be used in tests/test).

429 - tests/large contains several bash/python scripts that create random factor

430 graphs, compare several approximate inference algorithms (using tests/test) and

431 allow for easy visualization of the results using PyX.

432 - Added several .fg files for test purposes to /tests (e.g. two ALARM versions

433 alarm.fg and alarm_bnt.fg; testfast.fg, a 4x4 periodic Ising grid for

434 regression testing).

435 - Added a regression test to the Makefile which is included in the standard

436 target. It compares all inference methods on tests/testfast.fg with the

437 results stored in tests/testfast.out

439 Miscellaneous

441 - Expanded all tabs to spaces (":set tabstop 4\n:set expandtab\n:retab" in vim)

442 - Experimental MATLAB code added for StarEP approximation on cavity

443 - Renamed project to libDAI and changed directory name accordingly.

444 - Renamed JunctionTree to JTree.

445 - Fixed licensing (now it's officially GPL).

446 - Improved README

449 revision 252

450 ------------

452 Functionality

453 - Added RegionGraph, GBP, CVM and HAK (double-loop).

454 - Added JunctionTree (with two update algorithms, HUGIN and Shafer-Shenoy), which is a

455 RegionGraph.

456 - NormType is now chosen automatically (in case of negative factors, Prob::NORMLINF is used,

457 otherwise the default Prob::NORMPROB is used). Also, in case of negative factors, the

458 RegionGraph constructors assign each Factor to a unique outer region instead of dividing

459 it over all subsuming outer regions. See README for when negative factors are known to work

460 and when not.

461 - FactorGraph::FactorGraph(const vector<Factor>) only gives a warning in case of short loops,

462 it does not automatically merge factors anymore.

463 - Removed BP_SEQMAXRESNOCLEAR (all cavity initialization methods now are implicitly NOCLEAR)

464 - Added MATLAB interface functions ai_readfg, ai_removeshortloops and ai_bp

465 - Added LCBP-III type that should be equivalent to LCBP-II, but can handle zeroes

466 in potentials. Note that it is significantly slower than LCBP-II (and has to be reimplemented

467 such that it does not store the complete pancakes, but represents them as little factor graphs).

469 Implementation / code

470 - Made a seperate type WeightedGraph, which until now only implements Prim's

471 maximal spanning tree algorithm and is only used by the junction tree code. It might

472 be good to make it a class somewhere in the future and extend it's interface.

473 - Made a seperate class ClusterGraph, which is only used by the junction tree

474 code. It's main purpose is a graph-theoretical variable elimination algorithm.

475 - Implemented the heuristic "minimum-new-edges-in-clique-graph" for variable elimination.

476 - Massive code cleanup, moving towards "generic" programming style, using

477 multiple inheritance and polymorphism.

478 o BP, LCBP, MF, HAK and JunctionTree now inherit from a common DAIAlg class

479 o Made generic functions Marginal, Marginal2ndO, calcCavityDist, calcCavityDist2ndO, clamp

480 that can be used by FactorGraph-based DAIAlgs.

481 o Created TProb<T> class, which stores a probability vector (without the accompanying indexing

482 and VarSet) and provides functionality for it (which is used by TFactor<T>).

483 o Rewrote the VarSet class. It now caches its statespace(). It now privately inherits from set<Var>.

484 I had to overload the insert methods of set<Var> so that they calculate the new statespace.

485 o Rewrote the TFactor class. The TFactor class now HAS a TProb and HAS a VarSet.

486 - Rewrote BP to use the new TProb<T> interface. Performance of BP improved up to a factor 6 by:

487 o using Prob's instead of Factor's;

488 o splitting the multiplication of the messages into two parts (thanks to Vicenc!);

489 o optimizing the calculation of the beliefs (only the message calculations were optimized till now).

490 o replacing FactorGraph::_nb1 and _nb2 (which were set<size_t>) by vector<size_t>

491 - LCBP now seperately stores cavitydists and pancakes. Added InitPancakes() method

492 that takes the cavitydists and multiplies them with the relevant factors. This

493 resulted in an API change in AI which now accepts and returns initial cavitydists

494 instead of initial pancakes.

496 Minor changes

497 - Started writing DoxyGen documentation

498 - Renamed lcptab2fg.m matlab/ai_writefg.m

499 - Moved all matlab stuff to matlab/

500 - More detailed reporting (also reports clocks used).

501 - Marginal and Marginal2ndO now use *differences* in logZ to avoid NaNs.

502 - Improved testing suite.

503 - Removed logreal support.

504 - FactorGraph now also supports input streams and ignores comment lines in .fg files.

505 - Added tests/create_full_fg.cpp and tests/create_ising_fg.cpp which create

506 full and periodic 2D Ising networks according to some command line parameters.

507 - Now logZ really returns logZ instead of -logZ.

508 - Added FactorGraph::WriteToDotFile

511 0.1.4 (2006-04-13)

512 ------------------

513 - Added file IO routines to read and write factorgraphs.

514 - Added L-infinity normalization for use with (partially) negative factors.

515 - Renamed BetheF, MFF to logZ, which now use complex numbers to be able to

516 handle (partially) negative factors.

517 - Added test suite.

518 - All probabilities are now represented using double instead of LogReal<double>.

519 - Combined Alg and Alg3 into LCBP. Several update schemes possible.

520 - Combined several variants of BP into doBP. Several update schemes possible.

521 Now uses precalculated indices for optimization.

522 - Renamed Node -> Var and Potential -> Factor.

523 - Extensive code cleanup. More use of OO techniques. Changed API.

524 - MaxIter now means maximum number of passes (corresponding to number of

525 _parallel_ updates).

526 - MaxDiff now means the maximum L-infinity distance between the updated and

527 original single variable beliefs, for all AI methods.

528 - Implemented RBP which converges faster than BP for difficult problems.

529 - Now uses method parameter which is a bitmask combining outmethod and inmethod

530 (see ai.h).

533 0.1.3 (2006-03-22)

534 --------------------

535 - All AI methods now return maxdiff

536 - ai.cpp:

537 o Now returns maxdiffinner and maxdiffouter

538 o New BP2ndO innermethod (estimate only 2nd order cavity interactions)

539 o New InitCav outermethod (only create initial cavity distributions)

540 - bp.cpp:

541 o New CavityDist2ndO which estimates 2nd order cavity interactions

542 - Makefile:

543 o Bugfix: removed dependencies on algwim.*

546 0.1.2 (2006-02-28)

547 --------------------

548 - Cleaned up alg.cpp (removed Alg2 and its corresponding data structures).

549 - Added the possibility to provide initial cavity distributions as an input

550 argument to ai (not much error checking is done, so be careful).

551 - Potentials2mx now correctly sets the dimensions of the P field (i.e. for

552 the output arguments Q, Q0 of ai).

553 - Removed algwim.* since it does not work.

556 0.1.1 (2006-02-28)

557 --------------------

558 - The constructors of (Log)FactorGraph and LogFactorGraph from a

559 vector<(Log)Potential> now merge potentials to prevent short loops (of length

560 4) in the factor graph. These are used in ai to construct the factor graphs

561 from the psi argument. If compiled with DEBUG defined, the method calc_nb()

562 of BipGraph checks for the existence of short loops.

563 - Changed calling syntax of ai (now the actual syntax *does* correspond to its

564 description in the help).

565 - ai does not hook cout anymore (which caused weird segfaults).

566 - Fixed a bug in an assert statement in the matlab interface code in ai.cpp.

567 - Removed network.* since it is not useful.

570 0.1.0 (2006-02-28)

571 --------------------

572 First version worthy a version number.