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

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

3 distribution if no state is allowed

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

5 alias substitutions

6 * Interface changes:

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

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

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

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

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

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

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

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

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

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

17 operators

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

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

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

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

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

23 variables with the same labels

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

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

26 FactorGraph::end)

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

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

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

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

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

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

33 * Fixed bug in calcMarginal, calcPairBeliefs, calcPairBeliefsNew where

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

35 errors (thanks to Dan Preston for reporting this)

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

37 * Exception objects now remember their error code

38 * Added examples/example_springler.cpp

41 git 065eae35cbfcc36f1a945ae3053c80c23f366306

42 --------------------------------------------

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

44 * [Frederik Eaton] Added Gibbs sampler to algorithms

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

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

47 Merged TODO and FILEFORMAT into doxygen documentation

48 * examples/

49 - Moved example.cpp to examples/

50 - Added examples/example_bipgraph.cpp

51 - Added examples/example_varset.cpp

52 * Interface changes:

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

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

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

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

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

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

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

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

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

62 * Miscellaneous small changes and cleanups:

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

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

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

66 - Cleanup of matlab interface code

67 - Small improvement of utils/fginfo

68 - Small cleanup of BP code

69 - Switched Makefile.win to GNU Make syntax

72 libDAI-0.2.2 (2008-09-30)

73 -------------------------

75 New features:

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

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

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

79 * Added ExactInf class for brute force exact inference.

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

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

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

83 of a double is too small.

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

85 * Improved documentation.

87 Improved architecture:

88 * Added Exceptions framework.

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

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

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

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

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

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

95 made easy by using boost::foreach.

96 * Added conditional compilation of inference methods.

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

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

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

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

101 specialization with T = Var.

102 * Improved ClusterGraph implementation, yielding significant speedups

103 for the JunctionTree algorithm on large factorgraphs.

105 Code cleanup:

106 * Moved everything into namespace "dai".

107 * Renamed DEBUG to DAI_DEBUG to avoid conflicts.

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

109 * Removed utils/remove_short_loops and matlab/remove_short_loops.

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

111 * Improved index.h:

112 - Renamed Index -> IndexFor

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

114 25% speedup of testregression

115 - Replaced multind by Permute

116 - Added MultiFor

117 - Added State

118 * New funcstionality of factor.h.

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

120 inference algorithm, because the Properties framework was not as

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

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

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

124 entangled) code.

125 * Removed x2x.

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

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

129 Miscellaneous improvements:

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

131 GCC under cygwin.

132 * Contributions by Peter Gober:

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

134 * Misc contributions by Giuseppe Passino:

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

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

137 - changed #ifndefs to GNU style;

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

139 - dai::TProb:

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

141 o implementation of some methods via STL algorithms;

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

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

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

145 * Misc contributions by Christian Wojek:

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

147 and variables;

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

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

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

151 normtype is now Prob::NORMPROB by default.

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

154 Interface changes:

155 * VarSet::

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

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

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

159 - removed bool operator||

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

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

162 * FactorGraph::

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

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

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

166 - vars() -> vars

167 - factors() -> factors

168 - removed MakeFactorCavity(size_t)

169 - removed ExactMarginal(const VarSet &)

170 - removed ExactlogZ()

171 - removed updatedFactor(size_t)

172 - removed _normtype and NormType()

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

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

175 - undoProb(size_t) -> restoreFactor(size_t)

176 - saveProb(size_t) -> backupFactor(size_t)

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

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

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

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

181 - removed hasNegatives()

182 * RegionGraph::

183 - nr_ORs() -> nrORs()

184 - nr_IRs() -> nrIRs()

185 - ORs() -> ORs

186 - IRs() -> IRs

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

188 * Renamed Index -> IndexFor

189 * Diffs:

190 - max() -> maxDiff()

191 - max_size() -> maxSize()

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

193 * Factor::

194 - max() -> maxVal()

195 - part_sum() -> partSum()

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

197 * VarSet::operator&&

198 * Properties -> PropertySet

201 libDAI-0.2.1 (2008-05-26)

202 -------------------------

204 Bugfix release.

205 * added missing cstdio header in util.h

206 * fixed Properties in MR_CLAMPING_* and MR_EXACT_*

207 * added description of the factor graph fileformat

208 * improved Makefile

211 libDAI-0.2.0 (2006-11-30)

212 -------------------------

214 First public release.

217 0.1.5 (2006-11-30)

218 ------------------

220 Regressions

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

223 - EXACT method does not work anymore.

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

225 costs involved; inner loops should be optimized.

227 General framework

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

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

231 functions to forward functionality from either FactorGraph or RegionGraph

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

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

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

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

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

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

238 (make FactorGraph part of InfAlg).

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

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

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

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

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

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

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

246 saveProbs/undoProbs can give a significant speed increase.

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

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

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

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

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

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

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

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

255 command line (test) interface.

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

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

258 amount of code for handling enum parameters.

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

260 work for all approximate inference methods.

262 Bugfixes

264 - Fixed GBP free energy.

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

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

267 the logical NOT of what they should return).

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

269 - Fixed bug in utils/create_dreg_fg:

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

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

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

273 that consists of one outer region only.

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

275 - Fixed MaxDiffs flow (except for MR).

277 New functionality

279 - HAK supports several default cluster choices:

280 minimal (only factors)

281 delta (Markov blankets)

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

283 Only the maximal clusters are used as outer clusters.

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

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

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

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

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

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

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

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

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

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

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

295 It supports different initialization methods (the original RESPPROP,

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

297 (FULL and LINEAR).

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

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

300 marginals.

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

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

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

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

305 random factor graphs.

306 - Wrote utility to visualize factor graphs using graphviz.

307 (it uses the BOOST Program Options library)

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

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

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

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

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

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

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

315 - Added Martijn's x2x

317 Improvements of existing code

319 - Reimplemented RegionGraph and descendants: a RegionGraph ISA FactorGraph

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

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

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

323 outer regions from the factors.

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

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

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

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

328 timing and convergence criteria.

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

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

331 hopefully this will give more accurate approximations.

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

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

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

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

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

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

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

339 marginal functions to approximate cavity marginals.

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

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

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

343 arguments, randomize(), RemoveFirstOrderInteractions) and similar

344 operations to Prob

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

346 (tests/test is done).

347 - Renamed some FactorGraph methods:

348 nr_vars -> nrVars

349 nr_factors -> nrFactors

350 varind -> findVar

351 factorind -> findFactor

352 makeFacCavity -> makeFactorCavity

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

354 - Implemented RandomDRegularGraph

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

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

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

358 disconnected factor graphs).

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

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

361 - Removed calcCavityDist and calcCavityDist2ndO

362 - No longer depends on GSL.

363 - Increased portability by combining platform dependant utility functions

364 in util.{h,cpp}.

365 - Wrote *.m files providing help

367 Testing framework

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

370 functionality from the command line.

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

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

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

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

375 errors, relative logZ error and MaxDiff().

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

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

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

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

380 allow for easy visualization of the results using PyX.

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

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

383 regression testing).

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

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

386 results stored in tests/testfast.out

388 Miscellaneous

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

391 - Experimental MATLAB code added for StarEP approximation on cavity

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

393 - Renamed JunctionTree to JTree.

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

395 - Improved README

398 revision 252

399 ------------

401 Functionality

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

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

404 RegionGraph.

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

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

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

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

409 and when not.

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

411 it does not automatically merge factors anymore.

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

413 - Added MATLAB interface functions ai_readfg, ai_removeshortloops and ai_bp

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

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

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

418 Implementation / code

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

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

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

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

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

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

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

426 multiple inheritance and polymorphism.

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

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

429 that can be used by FactorGraph-based DAIAlgs.

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

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

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

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

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

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

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

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

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

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

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

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

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

443 instead of initial pancakes.

445 Minor changes

446 - Started writing DoxyGen documentation

447 - Renamed lcptab2fg.m matlab/ai_writefg.m

448 - Moved all matlab stuff to matlab/

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

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

451 - Improved testing suite.

452 - Removed logreal support.

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

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

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

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

457 - Added FactorGraph::WriteToDotFile

460 0.1.4 (2006-04-13)

461 ------------------

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

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

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

465 handle (partially) negative factors.

466 - Added test suite.

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

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

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

470 Now uses precalculated indices for optimization.

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

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

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

474 _parallel_ updates).

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

476 original single variable beliefs, for all AI methods.

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

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

479 (see ai.h).

482 0.1.3 (2006-03-22)

483 --------------------

484 - All AI methods now return maxdiff

485 - ai.cpp:

486 o Now returns maxdiffinner and maxdiffouter

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

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

489 - bp.cpp:

490 o New CavityDist2ndO which estimates 2nd order cavity interactions

491 - Makefile:

492 o Bugfix: removed dependencies on algwim.*

495 0.1.2 (2006-02-28)

496 --------------------

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

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

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

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

501 the output arguments Q, Q0 of ai).

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

505 0.1.1 (2006-02-28)

506 --------------------

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

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

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

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

511 of BipGraph checks for the existence of short loops.

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

513 description in the help).

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

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

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

519 0.1.0 (2006-02-28)

520 --------------------

521 First version worthy a version number.