Commit Graph

2417 Commits

Author SHA1 Message Date
Vitor Santos Costa
abf9691282 fixes to hashing code. 2008-11-18 11:29:08 +00:00
Vitor Santos Costa
99c5cb3e36 new version of term_hash based on SWI-ideas.
It handles infinite terms, and it seems to be a better hash.
2008-11-18 11:28:11 +00:00
Vitor Santos Costa
1b98de440d Improve Gibbs learning in CLP(BN). 2008-11-14 14:56:18 +00:00
Vitor Santos Costa
1be78e23ec had shortcut for var-to-var unification. 2008-11-14 14:54:52 +00:00
Vitor Santos Costa
e62ae6be36 improve handling of local stack expansion. 2008-11-14 14:54:25 +00:00
Vitor Santos Costa
f5e1b6d029 fix bug in create array: argument order was reversed. 2008-11-14 14:53:44 +00:00
Vitor Santos Costa
0ef147459a handle case where we can expand attributed variable stack
without having to grow stack physically.
2008-11-14 14:52:55 +00:00
Vitor Santos Costa
9296762791 fix bug in reserving space for queues. 2008-11-14 14:52:00 +00:00
Vitor Santos Costa
7a21ccb009 Include hack to unify two attributed variables, if the solver knows what
it is doing.
2008-11-14 14:50:54 +00:00
Vitor Santos Costa
8e9fde1c6c backtrack global hash table. 2008-11-14 14:50:04 +00:00
Vitor Santos Costa
99248795c9 support for learning CLP(BN) programs from Aleph 2008-11-13 09:03:53 +00:00
Vitor Santos Costa
1c334ea198 fix bad marking of attributed variables from trail. 2008-11-13 09:03:27 +00:00
Vitor Santos Costa
6920041189 fix bad size in grow stack 2008-11-13 09:02:33 +00:00
Vitor Santos Costa
297ea16dcb fix type in gprof 2008-11-05 13:28:44 +00:00
Vitor Santos Costa
3f491b8666 small fixes 2008-11-04 17:02:26 +00:00
Vitor Santos Costa
e19a8eeb87 fix case where index has no clauses. 2008-11-04 17:01:57 +00:00
Vitor Santos Costa
13fb5d5156 fix underflows when computing marginals. 2008-11-04 03:55:49 +00:00
Vitor Santos Costa
0dc7d3492d move aggregate support to aggregates. 2008-11-04 03:32:59 +00:00
Vitor Santos Costa
e19a7e6b61 include aggregate preparation in here. 2008-11-04 03:31:50 +00:00
Vitor Santos Costa
c81bc96fd0 fix EM support 2008-11-04 03:30:12 +00:00
Vitor Santos Costa
43365003dc support for EM with JT
use vel as default.
2008-11-04 03:29:11 +00:00
Vitor Santos Costa
8a9d66d001 more fixes to improve aggregate performance:
handle evidence on children and root nodes.
make graphviz call on top-level.
2008-11-04 00:46:17 +00:00
Vitor Santos Costa
f6c5d16f63 use avg CPT type.
how to deal with it is a solver problem, not an app issue.
2008-11-03 16:02:15 +00:00
Vitor Santos Costa
45df10e86d use avg cpt type. 2008-11-03 16:01:39 +00:00
Vitor Santos Costa
8938666bf5 add sum_list/3 2008-11-03 16:00:22 +00:00
Vitor Santos Costa
259313b603 fix the case when you refer a variable that does not affect the marginals. 2008-11-02 16:00:46 +00:00
Vitor Santos Costa
4490aa4bb6 predicate was not defined, so it should not be exported. 2008-11-02 16:00:19 +00:00
Vitor Santos Costa
53b8aa471c just small changes of paramters. 2008-11-02 15:59:41 +00:00
Vitor Santos Costa
941d3c3f13 EM was not checking for convergence properly. 2008-11-02 15:58:29 +00:00
Vitor Santos Costa
ac34be01b0 support for randomising or making uniform all distributions before starting EM 2008-11-01 21:09:08 +00:00
Vitor Santos Costa
4f1d7d0524 make this example a more uniform example. 2008-11-01 21:08:47 +00:00
Vitor Santos Costa
11e02be540 fixes to work with gibbs and vel. 2008-11-01 11:52:54 +00:00
Vitor Santos Costa
de79f30e45 use influences
fixes to work with EM.
2008-11-01 11:52:10 +00:00
Vitor Santos Costa
c6bb7e9dfe small interface fixes, use influences 2008-11-01 11:51:36 +00:00
Vitor Santos Costa
0ff1552370 normalise_CPT actually works on joints. 2008-11-01 11:49:38 +00:00
Vitor Santos Costa
ccbd6aea5b add new interface function to obtain key for a distribution 2008-11-01 11:48:53 +00:00
Vitor Santos Costa
4cebf98f0c fix bad variable passing
separate graph initilialisaton phase so that this can be done only once.
2008-11-01 11:47:56 +00:00
Vitor Santos Costa
08a736fea2 improve EM solver interface. 2008-11-01 11:47:17 +00:00
Vitor Santos Costa
c968576cb4 install Markov blanket analyser. 2008-11-01 11:46:08 +00:00
Vitor Santos Costa
4a2aa23c5b more CLP(BN) fixes, add softening option when computing MLE estimators. 2008-10-31 15:11:27 +00:00
Vitor Santos Costa
61f35779a0 make sure we include every vertex. 2008-10-31 14:10:57 +00:00
Vitor Santos Costa
9c1ee7ba34 add matrix_row to get a row of a matrix
fix a bit of error handling.
2008-10-31 09:47:58 +00:00
Vitor Santos Costa
aca06e16fc fix bad call to map_tree (obs from Paulo Moura). 2008-10-31 09:42:32 +00:00
Vitor Santos Costa
e66e7c86bc more fixes to use matrices in gibbs sampling. 2008-10-31 09:41:52 +00:00
Vitor Santos Costa
6408ae52ac we do not always want logarithms for CPTs. 2008-10-31 09:41:02 +00:00
Vitor Santos Costa
1d4dc2b0d7 we do not always want logarithms for CPTs. 2008-10-31 09:40:05 +00:00
Vitor Santos Costa
365f454755 use standard library for topsort. 2008-10-31 09:39:26 +00:00
Vitor Santos Costa
20e08fb549 find the variables connected with a variable, and all variables that may
affect each other.
2008-10-29 20:55:32 +00:00
Vitor Santos Costa
fa0d580faa make sure that we always consider evidence.
fix interface.
2008-10-29 20:54:15 +00:00
Vitor Santos Costa
d1db0d68ea distribution is now named through its structure, not just the main functor. 2008-10-29 20:52:56 +00:00