Fixed testem failure caused by rounding error
[libdai.git] / ChangeLog
1 * Added work-around for bug in Boost Graph Library
2 * [Frederik Eaton] Added Gibbs::state() accessors/mutators
3 * [Frederik Eaton] Fixed Gibbs::getVarDist(size_t) to return uniform
4 distribution if no state is allowed
5 * [Frederik Eaton] Improved parsing code in tests/testdai to allow recursive
6 alias substitutions
7 * Interface changes:
8 TProb<T>::minVal() -> TProb<T>::min()
9 TFactor<T>::minVal() -> TFactor<T>::min()
10 TProb<T>::maxVal() -> TProb<T>::max()
11 TFactor<T>::maxVal() -> TFactor<T>::max()
12 TProb<T>::totalSum() -> TProb<T>sum()
13 TFactor<T>::totalSum() -> TFactor<T>::sum()
14 * [Frederik Eaton] Added methods setUniform (), sumAbs(), argmax() to TProb<T>
15 * [Frederik Eaton] Added TAGS, lib targets to Makefile
16 * [Frederik Eaton] Add useful aliases to aliases.conf
17 * [Frederik Eaton] Removed unnecessary copy constructors and assignment
18 operators
19 * [Frederik Eaton] Change cout to cerr in warnings and error messages
20 * [Giuseppe Passino] Optimised maximum-residual BP by using a reversed ordered
21 set instead of the linear search (which can yield enormous speedups - a
22 factor 500 has been measured on a binary Ising grid with 128x128 variables!)
23 * Added debug assertions to Var which check for inconsistent dimensions of
24 variables with the same labels
25 * [Giuseppe Passino] Added prefix ++ operator to State (State::operator++())
26 * [Giuseppe Passino] Added iterators to FactorGraph (FactorGraph::begin,
27 FactorGraph::end)
28 * [Giuseppe Passino] Added iterators to TFactor (TFactor::begin, TFactor::end)
29 * [Giuseppe Passino] Added iterators to TProb (TProb::begin, TProb::end)
30 * [Giuseppe Passino] Added BP::findMaximum(), which can be used after running
31 max-product BP to construct a global state with maximum probability
32 * Improved Makefile (libDAI now also builds out-of-the-box on MacOSX,
33 thanks to Dan Preston; merged Makefile.win and Makefile.shared into Makefile)
34 * Fixed bug in calcMarginal, calcPairBeliefs, calcPairBeliefsNew where
35 states with zero probability mass were clamped, leading to NaNs or assertion
36 errors (thanks to Dan Preston for reporting this)
37 * toc() now returns POSIX system time with maximum accuracy of microseconds
38 * Exception objects now remember their error code
39 * Added examples/example_sprinkler.cpp
40
41
42 git 065eae35cbfcc36f1a945ae3053c80c23f366306
43 --------------------------------------------
44 * [Peter Gober] Made libDAI build out-of-the-box on Cygwin.
45 * [Frederik Eaton] Added Gibbs sampler to algorithms
46 * Improved documentation of include/dai/{bipgraph.h, prob.h, smallset.h,
47 var.h, varset.h, factor.h, enum.h} and of examples/example.cpp
48 Merged TODO and FILEFORMAT into doxygen documentation
49 * examples/
50 - Moved example.cpp to examples/
51 - Added examples/example_bipgraph.cpp
52 - Added examples/example_varset.cpp
53 * Interface changes:
54 - Replaced TProb<T>::log0() by TProb<T>::log(true)
55 - Replaced TProb<T>::takeLog0() by TProb<T>::takeLog(true)
56 - Replaced TFactor<T>::log0() by TFactor<T>::log(true)
57 - Removed TProb<T>::hasNonPositives()
58 - Renamed smallSet<T> to SmallSet<T>
59 - Removed TFactor<T>::divided_by, use operator/ instead
60 - Removed TFactor<T>::divide, use operator/= instead
61 - Removed TFactor<T>::partSum(const VarSet&), use marginal(const VarSet&,true) instead
62 - Improved constructors of TProb and TFactor to use iterators instead of pointers
63 * Miscellaneous small changes and cleanups:
64 - Changed regression test so that it also works under Windows
65 - Changed output stream operator<< for Var and VarSet
66 - Added TProb::draw() function, which draws a random index
67 - Cleanup of matlab interface code
68 - Small improvement of utils/fginfo
69 - Small cleanup of BP code
70 - Switched Makefile.win to GNU Make syntax
71
72
73 libDAI-0.2.2 (2008-09-30)
74 -------------------------
75
76 New features:
77 * Approximate inference methods now report the number of iterations needed.
78 * Added damping to various algorithms to improve convergence properties.
79 * Added more features to utils/createfg for creating factor graphs.
80 * Added ExactInf class for brute force exact inference.
81 * [Giuseppe Pasino] Added "logdomain" property to BP, a boolean that controls
82 whether calculations are done in the log-domain or in the linear domain;
83 doing calculations in the log-domain may help if the numerical range
84 of a double is too small.
85 * [Claudio Lima] Added Max-Product functionality to BP.
86 * Improved documentation.
87
88 Improved architecture:
89 * Added Exceptions framework.
90 * Pervasive change of BipartiteGraph implementation (based on an idea by
91 Giuseppe Passino). BipartiteGraph no longer stores the node properties
92 (former _V1 and _V2), nor does it store a dense adjacency matrix anymore,
93 nor an edge list. Instead, it stores the graph structure as lists of
94 neighboring nodes. This yields a significant memory/speed improvement for
95 large factor graphs, and is more elegant as well. Iterating over neighbors is
96 made easy by using boost::foreach.
97 * Added conditional compilation of inference methods.
98 * VarSet is now implemented using a std::vector<Var> instead of a
99 std::set<Var>, which yields a significant speed improvement. Furthermore,
100 the implementation has been generalized, resulting in the small_set<T> class
101 which can be used to represent sets of small cardinality; VarSet is the
102 specialization with T = Var.
103 * Improved ClusterGraph implementation, yielding significant speedups
104 for the JunctionTree algorithm on large factorgraphs.
105
106 Code cleanup:
107 * Moved everything into namespace "dai".
108 * Renamed DEBUG to DAI_DEBUG to avoid conflicts.
109 * Replaced ENUM2,ENUM3,ENUM4,ENUM5,ENUM6 by single DAI_ENUM macro.
110 * Removed utils/remove_short_loops and matlab/remove_short_loops.
111 * Replaced sub_nb class in mr.h by boost::dynamic_bitset.
112 * Improved index.h:
113 - Renamed Index -> IndexFor
114 - Added some .reserve()'s to IndexFor methods which yields a
115 25% speedup of testregression
116 - Replaced multind by Permute
117 - Added MultiFor
118 - Added State
119 * New funcstionality of factor.h.
120 * Moved Properties and MaxDiff frameworks from InfAlg to each individual
121 inference algorithm, because the Properties framework was not as
122 convenient as I hoped, and not every inference algorithm needs a maxdiff
123 variable. Also, replaced some FactorGraph functionality in InfAlg by a
124 function that returns the FactorGraph. The result is cleaner (less
125 entangled) code.
126 * Removed x2x.
127 * Replaced Complex with real numbers (negative potentials are just too rare
128 to warrant the additional "complexity" :)).
129
130 Miscellaneous improvements:
131 * Now compiles also with MS Visual C++ (thanks to Jiuxiang Hu) and with
132 GCC under cygwin.
133 * Contributions by Peter Gober:
134 - Renamed variable _N in mr.* for compatibility with g++ under cygwin.
135 * Misc contributions by Giuseppe Passino:
136 - removed "using namespace std;" from header files - bad practice;
137 - moved header files in include/dai and sources in src;
138 - changed #ifndefs to GNU style;
139 - added extra warning checks (-W -Wextra) and fixed resulting warnings;
140 - dai::TProb:
141 o removed copy constructor and assignment operators (redundant);
142 o implementation of some methods via STL algorithms;
143 o added methods takeExp, takeLog, takeLog0 for transformation in-place;
144 o explicit constructor (prevents implicit conversion from size_t to TProb);
145 o added operator+,+=,-,-=, with argument T (for calculations in log-scale);
146 * Misc contributions by Christian Wojek:
147 - New FactorGraph constructor that constructs from given ranges of factors
148 and variables;
149 - Optimization of FactorGraph constructors using tr1::unordered_map.
150 * FactorGraph constructors no longer check for short loops (huge speed
151 increase for large factor graphs), nor for negative entries. Also, the
152 normtype is now Prob::NORMPROB by default.
153 * Improved MaxSpanningTreePrims algorithm (uses boost::graph).
154
155 Interface changes:
156 * VarSet::
157 - VarSet::stateSpace() -> nrStates(const VarSet &)
158 - VarSet( const std::set<Var> ) -> VarSet( begin, end, sizeHint=0 )
159 - VarSet( const std::vector<Var> ) -> VarSet( begin, end, sizeHint=0 )
160 - removed bool operator||
161 - operator&&(const VarSet&) -> intersects(const VarSet&)
162 - operator&&(const Var&) -> contains(const Var&)
163 * FactorGraph::
164 - delta(const Var &) -> delta(size_t)
165 - Delta(const Var &) -> Delta(size_t)
166 - makeCavity(const Var &) -> makeCavity(size_t)
167 - vars() -> vars
168 - factors() -> factors
169 - removed MakeFactorCavity(size_t)
170 - removed ExactMarginal(const VarSet &)
171 - removed ExactlogZ()
172 - removed updatedFactor(size_t)
173 - removed _normtype and NormType()
174 - removed hasShortLoops(...) and removeShortLoops(...)
175 - WriteToDotFile(const char *filename) -> printDot( std::ostream& os )
176 - undoProb(size_t) -> restoreFactor(size_t)
177 - saveProb(size_t) -> backupFactor(size_t)
178 - undoProbs(const VarSet &) -> restoreFactors(const VarSet &)
179 - saveProbs(const VarSet &) -> backupFactors(const VarSet &)
180 - ReadFromFile(const char*) returns void (throws on error)
181 - WriteToFile(const char*) returns void (throws on error)
182 - removed hasNegatives()
183 * RegionGraph::
184 - nr_ORs() -> nrORs()
185 - nr_IRs() -> nrIRs()
186 - ORs() -> ORs
187 - IRs() -> IRs
188 * *::Regenerate() -> *::construct()
189 * Renamed Index -> IndexFor
190 * Diffs:
191 - max() -> maxDiff()
192 - max_size() -> maxSize()
193 * Prob::max() -> Prob::maxVal()
194 * Factor::
195 - max() -> maxVal()
196 - part_sum() -> partSum()
197 * toc() in util.h now returns seconds as a double
198 * VarSet::operator&&
199 * Properties -> PropertySet
200
201
202 libDAI-0.2.1 (2008-05-26)
203 -------------------------
204
205 Bugfix release.
206 * added missing cstdio header in util.h
207 * fixed Properties in MR_CLAMPING_* and MR_EXACT_*
208 * added description of the factor graph fileformat
209 * improved Makefile
210
211
212 libDAI-0.2.0 (2006-11-30)
213 -------------------------
214
215 First public release.
216
217
218 0.1.5 (2006-11-30)
219 ------------------
220
221 Regressions
222
223 - tests/testlcbp and tests/testlcbp are broken.
224 - EXACT method does not work anymore.
225 - The Properties framework gives a speed penalty because of the lookup
226 costs involved; inner loops should be optimized.
227
228 General framework
229
230 - DAIAlg is now a template class; typedefs for DAIAlg<FactorGraph> and for
231 DAIAlg<RegionGraph> are provided. In this way, we do not have to write "wrapper"
232 functions to forward functionality from either FactorGraph or RegionGraph
233 to DAIAlg. Functionality like clamping can be implemented in FactorGraph
234 and in RegionGraph and no explicit interface is needed in descendants.
235 - New abstract base class InfAlg added, representing an inference algorithm,
236 from which DAIAlg<T> inherits. This solves the incompatibility problems of
237 DAIAlg<T> for different T (e.g. DAIAlg<FactorGraph> was incompatible with
238 DAIAlg<RegionGraph>). More work is required to reduce code duplication
239 (make FactorGraph part of InfAlg).
240 - Added generic interface (nrVars(), Vars(), nrFactors(), factor(size_t),
241 beliefs(), belief(VarSet &), ...) to InfAlg and descendants.
242 - Added a saveProbs/undoProbs interface to InfAlg and descendants that enables
243 one to save a few factors, modify them (e.g. clamp them), and then restore them
244 to their old values. Undo should also init the corresponding messages / beliefs.
245 This can be used if a given factor graph repeatedly needs to be clamped in
246 different ways and an approximation method is run for each clamping; using the
247 saveProbs/undoProbs can give a significant speed increase.
248 - Switched to a general Properties framework that handles the parameters of
249 all inference methods in a uniform manner. The Properties class is a map of
250 several properties in boost::any objects, indexed by their names (strings).
251 It can read from a stream and write to a stream. It is recursive, in the sense
252 that a Properties object can hold a variable of type Properties as well.
253 - Added a generic way of constructing inference algorithms from a factor graph,
254 name and properties object. Added the newInfAlg function which constructs
255 the requested object. This is used by LCBP, the Matlab interface and the
256 command line (test) interface.
257 - Added a generic enum framework for enum parameters. Although implemented as a
258 hack, it is quite useful in that it drastically simplifies and reduces the
259 amount of code for handling enum parameters.
260 - Provided generic functions for calculating marginals in different ways that
261 work for all approximate inference methods.
262
263 Bugfixes
264
265 - Fixed GBP free energy.
266 - Fixed bug in junctiontree (it didn't init the _vars variable).
267 - Corrected two bugs in operator&& and operator|| in VarSet (they returned
268 the logical NOT of what they should return).
269 - Fixed bug in RegionGraph::RecomputeOR(s).
270 - Fixed bug in utils/create_dreg_fg:
271 graph structure was not random for given parameters (forgot to call srand()).
272 - TreeEP bug workaround: use the complete junction tree instead of a subtree.
273 - Fixed bug in JTree::HUGIN() and JTree:ShaferShenoy() in case of junction tree
274 that consists of one outer region only.
275 - Fixed INIT bug in LCBP2::UpdatePancake().
276 - Fixed MaxDiffs flow (except for MR).
277
278 New functionality
279
280 - HAK supports several default cluster choices:
281 minimal (only factors)
282 delta (Markov blankets)
283 loop (all loops consisting of loops consisting of <loopdepth> or less variables)
284 Only the maximal clusters are used as outer clusters.
285 - Implemented TreeEP. It generalizes the heuristic method described in the
286 Minka & Qi paper for obtaining a tree with the most relevant interactions to
287 higher order interactions. Almost all optimizations described in the Minka & Qi
288 paper are used, except that evidence is passed over the whole tree instead of
289 relevant subsets (the latter is almost implemented but buggy). Also added
290 alternative (worst-case) algorithm that uses a maximum spanning tree on the
291 weighted graph where the weight between neighbours i and j is given by
292 N(psi,i,j), where psi is the product of all the factors involving both i and j
293 (which is an upper bound on the effective interaction between i and j).
294 - Implemented MR (MontanariRizzo) based on Bastian's code, but extended it
295 to be able to handle connectivities larger than 3 (in principle, up to 32).
296 It supports different initialization methods (the original RESPPROP,
297 the CLAMPING method and EXACT which uses JTree) and different update methods
298 (FULL and LINEAR).
299 - Implemented LCBP2, an analogon of LCBP which represents pancakes as little
300 networks and uses some approximate inference method on them for calculating
301 marginals.
302 - Now there are several LCBP variants (LCBP, LCBPI, LCBPJ, LCBPK, LCBPL);
303 LCBPJ works only for pairwise, LCBPK is LCBP improved for higher order
304 interactions and LCBPL is LCBPI improved for higher-order interactions.
305 - Wrote one single program utils/createfg for creating various types of
306 random factor graphs.
307 - Wrote utility to visualize factor graphs using graphviz.
308 (it uses the BOOST Program Options library)
309 - Added fginfo utility that displays some info about a .fg file.
310 - Implemented Factor::strength function that calculates the potential strength
311 N(psi,i,j) between variables i and j as described in cs.IT:0504030
312 - Wrote a general MatLab interface matlab/dai (similar to tests/test);
313 this unified the matlab functions dai, dai_bp, dai_mf, dai_jt, dai_tep, dai_cvm.
314 - Added MATLAB routine that returns contraction matrix A for BP convergence analysis.
315 - Implemented a MATLAB interface ai_potstrength for Factor::strength
316 - Added Martijn's x2x
317
318 Improvements of existing code
319
320 - Reimplemented RegionGraph and descendants: a RegionGraph ISA FactorGraph
321 and also a BipartiteGraph<FRegion,Region>. It now also keeps a map that
322 associates outer region indices to factor indices (no powers yet, this
323 is deemed superfluous) and provides functions to recompute (some of) the
324 outer regions from the factors.
325 - InfAlg descendants run() methods now stop immediately and return NAN in case
326 they detect NANs. Only BP does not do NAN checking for performance reasons.
327 - LCBP now works with factors containing zeroes (by defining x/0 = 0).
328 - HAK, GBP and DoubleLoop now conform to the standards for verbose reporting,
329 timing and convergence criteria.
330 - Implemented logZ() for JTree. It does the calculation during message-passing.
331 - Marginal2ndO now optionally divides by the single node beliefs (to the power n-2);
332 hopefully this will give more accurate approximations.
333 - Marginal and Marginal2ndO (optionally) use the new saveProbs/undoProbs functionality
334 for a faster way of calculating marginals, which does not require a call to init()
335 nor cloning the whole object for each clamping. This leads to a significant speedup.
336 - LCBP (and LCBP2) now have complete flexibility in the specification of the
337 inner method, i.e. the method used to generate the initial cavity approximations.
338 One can pass two strings, a name and a properties string, and LCBP simply invokes
339 newInfAlg to construct the corresponding inference algorithm and uses the generic
340 marginal functions to approximate cavity marginals.
341 - Replaced the global "method" variable by local properties and removed ai.h
342 - Added some methods to Factor (operators *, *=, /, /= with doubles as
343 second argument, operators -, +=, -= with other Factors as second
344 arguments, randomize(), RemoveFirstOrderInteractions) and similar
345 operations to Prob
346 - Moving towards boost::program_options for handling command line arguments
347 (tests/test is done).
348 - Renamed some FactorGraph methods:
349 nr_vars -> nrVars
350 nr_factors -> nrFactors
351 varind -> findVar
352 factorind -> findFactor
353 makeFacCavity -> makeFactorCavity
354 - LCBP_SEQMAXRES has been removed because it did strange things.
355 - Implemented RandomDRegularGraph
356 - Implemented JTree::calcMarginal for marginals that are not confined
357 within one cluster (using cut-set conditioning).
358 - Added isConnected() method to FactorGraph (some methods do not work with
359 disconnected factor graphs).
360 - Pair beliefs are now calculated in a symmetrical way by calcPairBeliefs
361 - Removed single node interaction "correction" code from clamping methods
362 - Removed calcCavityDist and calcCavityDist2ndO
363 - No longer depends on GSL.
364 - Increased portability by combining platform dependant utility functions
365 in util.{h,cpp}.
366 - Wrote *.m files providing help
367
368 Testing framework
369
370 - Made a new and significantly improved testing framework that provides most
371 functionality from the command line.
372 - The basis is provided by tests/test, which uses the newInfAlg functionality
373 and enables the user to easily compare from the command line different
374 inference methods on a given factorgraph. All parameters can be specified.
375 Output consists of CPU time, average and maximum single variable marginal
376 errors, relative logZ error and MaxDiff().
377 - tests/aliases.conf contains aliases for standard combinations of methods
378 and their options (which can be used in tests/test).
379 - tests/large contains several bash/python scripts that create random factor
380 graphs, compare several approximate inference algorithms (using tests/test) and
381 allow for easy visualization of the results using PyX.
382 - Added several .fg files for test purposes to /tests (e.g. two ALARM versions
383 alarm.fg and alarm_bnt.fg; testfast.fg, a 4x4 periodic Ising grid for
384 regression testing).
385 - Added a regression test to the Makefile which is included in the standard
386 target. It compares all inference methods on tests/testfast.fg with the
387 results stored in tests/testfast.out
388
389 Miscellaneous
390
391 - Expanded all tabs to spaces (":set tabstop 4\n:set expandtab\n:retab" in vim)
392 - Experimental MATLAB code added for StarEP approximation on cavity
393 - Renamed project to libDAI and changed directory name accordingly.
394 - Renamed JunctionTree to JTree.
395 - Fixed licensing (now it's officially GPL).
396 - Improved README
397
398
399 revision 252
400 ------------
401
402 Functionality
403 - Added RegionGraph, GBP, CVM and HAK (double-loop).
404 - Added JunctionTree (with two update algorithms, HUGIN and Shafer-Shenoy), which is a
405 RegionGraph.
406 - NormType is now chosen automatically (in case of negative factors, Prob::NORMLINF is used,
407 otherwise the default Prob::NORMPROB is used). Also, in case of negative factors, the
408 RegionGraph constructors assign each Factor to a unique outer region instead of dividing
409 it over all subsuming outer regions. See README for when negative factors are known to work
410 and when not.
411 - FactorGraph::FactorGraph(const vector<Factor>) only gives a warning in case of short loops,
412 it does not automatically merge factors anymore.
413 - Removed BP_SEQMAXRESNOCLEAR (all cavity initialization methods now are implicitly NOCLEAR)
414 - Added MATLAB interface functions ai_readfg, ai_removeshortloops and ai_bp
415 - Added LCBP-III type that should be equivalent to LCBP-II, but can handle zeroes
416 in potentials. Note that it is significantly slower than LCBP-II (and has to be reimplemented
417 such that it does not store the complete pancakes, but represents them as little factor graphs).
418
419 Implementation / code
420 - Made a seperate type WeightedGraph, which until now only implements Prim's
421 maximal spanning tree algorithm and is only used by the junction tree code. It might
422 be good to make it a class somewhere in the future and extend it's interface.
423 - Made a seperate class ClusterGraph, which is only used by the junction tree
424 code. It's main purpose is a graph-theoretical variable elimination algorithm.
425 - Implemented the heuristic "minimum-new-edges-in-clique-graph" for variable elimination.
426 - Massive code cleanup, moving towards "generic" programming style, using
427 multiple inheritance and polymorphism.
428 o BP, LCBP, MF, HAK and JunctionTree now inherit from a common DAIAlg class
429 o Made generic functions Marginal, Marginal2ndO, calcCavityDist, calcCavityDist2ndO, clamp
430 that can be used by FactorGraph-based DAIAlgs.
431 o Created TProb<T> class, which stores a probability vector (without the accompanying indexing
432 and VarSet) and provides functionality for it (which is used by TFactor<T>).
433 o Rewrote the VarSet class. It now caches its statespace(). It now privately inherits from set<Var>.
434 I had to overload the insert methods of set<Var> so that they calculate the new statespace.
435 o Rewrote the TFactor class. The TFactor class now HAS a TProb and HAS a VarSet.
436 - Rewrote BP to use the new TProb<T> interface. Performance of BP improved up to a factor 6 by:
437 o using Prob's instead of Factor's;
438 o splitting the multiplication of the messages into two parts (thanks to Vicenc!);
439 o optimizing the calculation of the beliefs (only the message calculations were optimized till now).
440 o replacing FactorGraph::_nb1 and _nb2 (which were set<size_t>) by vector<size_t>
441 - LCBP now seperately stores cavitydists and pancakes. Added InitPancakes() method
442 that takes the cavitydists and multiplies them with the relevant factors. This
443 resulted in an API change in AI which now accepts and returns initial cavitydists
444 instead of initial pancakes.
445
446 Minor changes
447 - Started writing DoxyGen documentation
448 - Renamed lcptab2fg.m matlab/ai_writefg.m
449 - Moved all matlab stuff to matlab/
450 - More detailed reporting (also reports clocks used).
451 - Marginal and Marginal2ndO now use *differences* in logZ to avoid NaNs.
452 - Improved testing suite.
453 - Removed logreal support.
454 - FactorGraph now also supports input streams and ignores comment lines in .fg files.
455 - Added tests/create_full_fg.cpp and tests/create_ising_fg.cpp which create
456 full and periodic 2D Ising networks according to some command line parameters.
457 - Now logZ really returns logZ instead of -logZ.
458 - Added FactorGraph::WriteToDotFile
459
460
461 0.1.4 (2006-04-13)
462 ------------------
463 - Added file IO routines to read and write factorgraphs.
464 - Added L-infinity normalization for use with (partially) negative factors.
465 - Renamed BetheF, MFF to logZ, which now use complex numbers to be able to
466 handle (partially) negative factors.
467 - Added test suite.
468 - All probabilities are now represented using double instead of LogReal<double>.
469 - Combined Alg and Alg3 into LCBP. Several update schemes possible.
470 - Combined several variants of BP into doBP. Several update schemes possible.
471 Now uses precalculated indices for optimization.
472 - Renamed Node -> Var and Potential -> Factor.
473 - Extensive code cleanup. More use of OO techniques. Changed API.
474 - MaxIter now means maximum number of passes (corresponding to number of
475 _parallel_ updates).
476 - MaxDiff now means the maximum L-infinity distance between the updated and
477 original single variable beliefs, for all AI methods.
478 - Implemented RBP which converges faster than BP for difficult problems.
479 - Now uses method parameter which is a bitmask combining outmethod and inmethod
480 (see ai.h).
481
482
483 0.1.3 (2006-03-22)
484 --------------------
485 - All AI methods now return maxdiff
486 - ai.cpp:
487 o Now returns maxdiffinner and maxdiffouter
488 o New BP2ndO innermethod (estimate only 2nd order cavity interactions)
489 o New InitCav outermethod (only create initial cavity distributions)
490 - bp.cpp:
491 o New CavityDist2ndO which estimates 2nd order cavity interactions
492 - Makefile:
493 o Bugfix: removed dependencies on algwim.*
494
495
496 0.1.2 (2006-02-28)
497 --------------------
498 - Cleaned up alg.cpp (removed Alg2 and its corresponding data structures).
499 - Added the possibility to provide initial cavity distributions as an input
500 argument to ai (not much error checking is done, so be careful).
501 - Potentials2mx now correctly sets the dimensions of the P field (i.e. for
502 the output arguments Q, Q0 of ai).
503 - Removed algwim.* since it does not work.
504
505
506 0.1.1 (2006-02-28)
507 --------------------
508 - The constructors of (Log)FactorGraph and LogFactorGraph from a
509 vector<(Log)Potential> now merge potentials to prevent short loops (of length
510 4) in the factor graph. These are used in ai to construct the factor graphs
511 from the psi argument. If compiled with DEBUG defined, the method calc_nb()
512 of BipGraph checks for the existence of short loops.
513 - Changed calling syntax of ai (now the actual syntax *does* correspond to its
514 description in the help).
515 - ai does not hook cout anymore (which caused weird segfaults).
516 - Fixed a bug in an assert statement in the matlab interface code in ai.cpp.
517 - Removed network.* since it is not useful.
518
519
520 0.1.0 (2006-02-28)
521 --------------------
522 First version worthy a version number.