dellaert
|
19a06ca68f
|
Added unit tests for NULL noise model and fixed bug in hessianDiagonalXX for that case
|
2014-02-16 00:21:15 -05:00 |
Luca
|
71e4015304
|
bug fix: before hessianDiagonal seg-faulted with no noise model
|
2014-02-15 16:44:20 -05:00 |
|
dellaert
|
a2829fffad
|
Working and tested implementation of hessianBlockDiagonal
|
2014-02-14 13:31:40 -05:00 |
|
dellaert
|
83918be8cd
|
Put in empty bodies for hessianBlockDiagonal
|
2014-02-14 13:18:59 -05:00 |
|
dellaert
|
4ce7d47bce
|
Fixed bug for non-unit noise models
|
2014-02-14 01:38:31 -05:00 |
|
dellaert
|
76959d4d18
|
Added hessianDiagonal; works for Gaussian factor graphs with JacobianFactors
|
2014-02-14 01:14:32 -05:00 |
Alex Cunningham
|
96a4d80da4
|
Fixed some warnings
|
2014-01-04 20:35:19 -05:00 |
Richard Roberts
|
2553e39746
|
Using an "Empty" constant in VariableSlots instead of calling numeric_limits::max() each time
|
2013-12-22 14:26:47 -05:00 |
|
Richard Roberts
|
09309e2869
|
Fixed debug and error message in JacobianFactor _countDims
|
2013-12-22 14:26:47 -05:00 |
|
Richard Roberts
|
411e934385
|
Fixed a bug in checking for empty-matrix Jacobians that still have keys (kind of a hack)
|
2013-11-05 16:06:14 +00:00 |
|
Richard Roberts
|
2dc40087d0
|
Fixed problem in ISAM2 marginalizeLeaves (with Luca, port from other repo)
|
2013-11-05 16:06:06 +00:00 |
Frank Dellaert
|
6c1f851d51
|
small change as alpha=1
|
2013-10-29 14:09:58 +00:00 |
|
Frank Dellaert
|
daf5938577
|
const correctness
|
2013-10-29 04:58:51 +00:00 |
|
Frank Dellaert
|
4591835223
|
gradientAtZero is now implemented in the fcators (Richard says, need to work on efficiency)
|
2013-10-29 03:42:33 +00:00 |
Pablo Fernandez Alcantarilla
|
621483bc9b
|
multiplyHessian in JacobianFactor works
|
2013-10-24 16:40:25 +00:00 |
|
Pablo Fernandez Alcantarilla
|
6a383799d7
|
Skeleton code for GaussianFactorGraph::multiplyHessian
|
2013-10-24 15:52:32 +00:00 |
|
Richard Roberts
|
95ac34aeeb
|
Updating Hessian in-place instead of computing Hessian for each Jacobian
|
2013-10-03 16:50:16 +00:00 |
|
Richard Roberts
|
bd89c5fd45
|
Using FastVector (with tbb allocator) instead of vector
|
2013-08-15 17:21:20 +00:00 |
|
Richard Roberts
|
3528173781
|
Fixed compile problem on linux using boost::join with boost::cref_list_of
|
2013-08-08 16:29:55 +00:00 |
|
Richard Roberts
|
2388f5df45
|
Revert "Fixing issues with boost::assign"
This reverts commit d820732adc34a6622aae5d2bc0eb4a88e133beaf.
|
2013-08-08 16:29:48 +00:00 |
|
Richard Roberts
|
6556fb5b1e
|
Fixed a couple errors and warnings
|
2013-08-07 21:57:25 +00:00 |
Alex Cunningham
|
2b1a533477
|
Fixing issues with boost::assign
|
2013-08-07 13:29:35 +00:00 |
|
Richard Roberts
|
7aa1c61dcd
|
Fixed problem is JacobianFactor combine with zero-row factors
|
2013-08-06 23:55:38 +00:00 |
|
Richard Roberts
|
375f7c16e6
|
Working on fixing ISAM
|
2013-08-06 19:56:48 +00:00 |
|
Richard Roberts
|
8ae920d85a
|
Fixed matrix/noise model sizes in JacobianFactor::splitConditional after constrained QR
|
2013-08-06 17:36:03 +00:00 |
|
Richard Roberts
|
e39d100b6a
|
Made most global unit tests compile, includes dogleg, iterative, kalman filter, etc
|
2013-08-06 13:44:22 +00:00 |
|
Richard Roberts
|
0e80fe6418
|
Moved ordering tests, fixed serialization and constructors, to make all inference, symbolic, and linear tests pass
|
2013-08-05 22:31:09 +00:00 |
|
Richard Roberts
|
1f6ca330ee
|
Cleanups
|
2013-08-02 22:09:30 +00:00 |
|
Richard Roberts
|
0d05bf5ac5
|
Started on HessianFactor, converted Nonlinear stuff and disabled a lot of more advanced classes so the library compiles with nonlinear using unordered.
|
2013-08-01 21:57:43 +00:00 |
|
Richard Roberts
|
3ade190128
|
Renamed old classes to "Ordered" and removed "Unordered" name from new classes
|
2013-07-29 23:55:40 +00:00 |
|
Richard Roberts
|
941cd93bf6
|
Returning empty noise model from QR and fix for backwards-compatibility in old code
|
2013-07-12 22:27:44 +00:00 |
|
Frank Dellaert
|
75751cc5fa
|
Comments
|
2013-06-24 15:33:39 +00:00 |
|
Richard Roberts
|
547323cc79
|
Fixed a bunch of compiler warnings
|
2013-04-05 21:34:04 +00:00 |
|
Richard Roberts
|
1755136b1b
|
Added JacobianFactor constructor from a GFG that merges the factors
|
2012-11-26 19:21:05 +00:00 |
|
Richard Roberts
|
eb21cf0911
|
Cleaned up information/augmentedInformation for GaussianFactor and GaussianConditional
|
2012-10-08 22:40:37 +00:00 |
|
Richard Roberts
|
89b50e7679
|
Renamed tic -> gttic and toc -> gttoc to avoid conflict with PCL tic/toc
|
2012-10-02 20:18:41 +00:00 |
|
Richard Roberts
|
4876cc7ff7
|
Revamped timing statements - much easier to use, exception-safe (see email to frankcvs list)
|
2012-10-02 18:36:39 +00:00 |
Chris Beall
|
4297d24c96
|
changed tabs to spaces for consistent indentation in all of GTSAM
|
2012-10-02 14:40:07 +00:00 |
|
Frank Dellaert
|
aeb43bc8fc
|
Used technique described in http://gcc.gnu.org/onlinedocs/gcc/Diagnostic-Pragmas.html to turn off excessive warnings generated by boost lambda headers
|
2012-09-16 18:07:50 +00:00 |
|
Richard Roberts
|
6d1b86c2e0
|
Moved JacobianFactor type check/conversion functions into SubgraphSolver instead of GaussianFactorGraph and JacobianFactor
|
2012-09-06 14:33:44 +00:00 |
Yong-Dian Jian
|
4443752a18
|
clean the gfg to jfg conversion code
|
2012-09-05 17:04:48 +00:00 |
|
Richard Roberts
|
3e93c488e5
|
Removed extra code that was sometimes maintaining a staircase pattern in JacobianFactor by sorting rows - a holdover from the staircase optimization that is no longer done now that we are using Eigen's QR
|
2012-08-27 22:29:56 +00:00 |
|
Richard Roberts
|
a5d7695fac
|
Removed commented-out code
|
2012-08-22 22:40:31 +00:00 |
|
Richard Roberts
|
dc884edceb
|
No longer should permute JacobianFactor to sort keys when converting from HessianFactor
|
2012-08-22 22:40:30 +00:00 |
|
Richard Roberts
|
1dbda3f7ed
|
Standardized and corrected error handling on the linear size - underconstrained and negative systems always throw IndeterminantLinearSystemException, and all assertions checking for infinite values are removed. Also, we were not properly checking the result of Eigen's Cholesky, so sometimes elimination continued with incorrect matrices despite being underconstrained when Cholesky failed but did not produce NaN's.
|
2012-08-22 22:40:27 +00:00 |
|
Alex Cunningham
|
9ee098b1c1
|
Added separate splitConditional() to remove the top part of a JacobianFactor after performing QR
|
2012-08-15 17:35:14 +00:00 |
|
Alex Cunningham
|
b2e15eea4e
|
Added negate() to GaussianFactor, which computes the Anti-factor eqiuvalent, using implementation from the nonlinear Anti-factor
|
2012-06-28 19:43:00 +00:00 |
|
Alex Cunningham
|
c7734db4fa
|
Added formatting objects to linear and inference objects to allow for printing meaningful keys
|
2012-06-25 21:19:38 +00:00 |
|
Alex Cunningham
|
9b473023e6
|
Fixed warnings
|
2012-06-23 18:12:15 +00:00 |
|
Frank Dellaert
|
617bf071cb
|
Fixed warning by making "what" const. Please investigate warnings!
|
2012-06-22 21:44:29 +00:00 |
|
Richard Roberts
|
685d4d7da2
|
Added check and exception for incompatible noise model dimension in JacobianFactor
|
2012-06-22 15:24:57 +00:00 |
|
Richard Roberts
|
d57fc32e74
|
To simplify FactorGraph, removed convertCastFactors and dynamicCastFactors from FactorGraph - replaced their calls with in-place code to do the needed conversions
|
2012-06-18 14:55:30 +00:00 |
|
Richard Roberts
|
97f52c5643
|
Cleaned up boost bind dependency
|
2012-06-07 16:12:55 +00:00 |
|
Yong-Dian Jian
|
a9c36fc172
|
move linear algebra functions for jacobian factor graph to a new file
|
2012-06-07 01:24:19 +00:00 |
|
Richard Roberts
|
6e2312294c
|
Added computeInformation function to GaussianFactor to properly compute information matrix including noise models, and using it to fix bug in Marginals where noise model was not being accounted for (only affects when hard constraints are used)
|
2012-05-23 20:56:22 +00:00 |
|
Richard Roberts
|
510e2eacac
|
More progress in compiling on windows
|
2012-05-22 22:52:17 +00:00 |
|
Richard Roberts
|
1ce95c1d89
|
Removed LDL in favor of Cholesky
|
2012-05-15 15:49:14 +00:00 |
|
Richard Roberts
|
91e7dc5882
|
Caching linearized factors in iSAM2, improves speed when linearization is expensive
Merge remote-tracking branch 'svn/branches/iSAM2_cache_linearized' into trunk
Conflicts:
.cproject
|
2012-04-11 14:17:59 +00:00 |
|
Richard Roberts
|
7243a9870b
|
Using vector::assign instead of erase in JacobianFactor::eliminate, caused invalid iterator on some STL implementations
|
2012-04-09 13:21:16 +00:00 |
|
Richard Roberts
|
22ebe16a31
|
Merge remote-tracking branch 'svn/branches/incremental_dogleg_points_to_merge' into trunk
Conflicts:
.cproject
gtsam/linear/GaussianBayesTree-inl.h
gtsam/linear/GaussianBayesTree.cpp
gtsam/linear/GaussianBayesTree.h
gtsam/nonlinear/DoglegOptimizerImpl.h
gtsam/nonlinear/GaussianISAM2-inl.h
gtsam/nonlinear/GaussianISAM2.cpp
gtsam/nonlinear/GaussianISAM2.h
gtsam/nonlinear/ISAM2-impl.cpp
gtsam/nonlinear/ISAM2-inl.h
gtsam/nonlinear/ISAM2.h
|
2012-03-23 04:31:54 +00:00 |
|
Richard Roberts
|
d577655f1b
|
Removed problematic 'inline'
|
2012-03-23 04:05:51 +00:00 |
|
Yong-Dian Jian
|
30557ce4a5
|
remove a small overhead
|
2012-03-15 17:23:01 +00:00 |
|
Richard Roberts
|
4137cfe9d7
|
Added missing copy constructor for GaussianConditional, and assignment operators for GaussianConditional, JacobianFactor, and HessianFactor that properly copy the block matrices (require calling a special function because they contain references)
|
2012-01-22 05:16:12 +00:00 |
|
Richard Roberts
|
099f170bf2
|
Better error message when trying to convert a negative HessianFactor to a JacobianFactor
|
2012-01-20 17:20:56 +00:00 |
|
Richard Roberts
|
ffd01c7a7d
|
Removed unneeded TODO's
|
2012-01-18 21:22:05 +00:00 |
|
Richard Roberts
|
4b95a2799b
|
Bug fix in converting HessianFactor to JacobianFactor - had to zero out lower triangle of eliminated matrix. Additional unit tests to catch this.
|
2012-01-18 20:44:27 +00:00 |
|
Alex Cunningham
|
927217ba4b
|
Cleanup from FactorGraph-inl.h
|
2011-12-21 16:59:37 +00:00 |
|
Richard Roberts
|
6e1136ba20
|
(in branch) Separate gradient functions for FactorGraph, BayesNet, BayesTree, and BayesTree with ISAM2 Cliques (specialized for using stored gradient contributions in ISAM2 cliques)
|
2011-12-13 22:51:28 +00:00 |
|
Frank Dellaert
|
58d83f704f
|
Documentation and headers
|
2011-10-29 04:00:14 +00:00 |
|
Richard Roberts
|
2c53df3ee7
|
Simplified VectorValues interface and implementation, slight simplification to Permuted<>
|
2011-10-26 02:04:06 +00:00 |
|
Frank Dellaert
|
fabde56642
|
Doxygen
|
2011-10-22 19:56:26 +00:00 |
|
Richard Roberts
|
63ca74e492
|
Fixed bug in multifrontal marginals caused by backwards permutations with LDL (seems to be an inconsistency in Eigen?). Added GaussianConditional constructor from any number of frontal and parent variables. Added several new unit tests on marginals. Fixed small bug in GaussianConditional non-const get_d_ and get_R_ functions that didn't account for multiple frontal variables.
|
2011-09-23 02:50:46 +00:00 |
|
Alex Cunningham
|
3c7836cf4d
|
comments only
|
2011-06-13 20:40:44 +00:00 |
|
Alex Cunningham
|
fa4faa274a
|
Merge branch 'prep_0.9.3'
|
2011-06-13 16:55:31 +00:00 |
|
Richard Roberts
|
5ea18829b5
|
Fixed code comments to indicate that assertInvariants does not check for sorted keys
|
2011-06-04 12:54:43 +00:00 |
|
Alex Cunningham
|
79c09708e8
|
Assorted cleanup to remove ublas references, switch more Vector/Matrix utility functions to use Eigen in templates, reimplemented backsubstitution with Eigen
|
2011-06-02 20:35:02 +00:00 |
|
Alex Cunningham
|
2e942f08ac
|
Removed some extraneous comments from eigen transition
|
2011-05-20 18:43:37 +00:00 |
|
Alex Cunningham
|
e20561be73
|
Merge branch 'eigen'
|
2011-05-20 13:52:08 +00:00 |
|
Frank Dellaert
|
e635b21154
|
made keys_ and nrFrontals_ private to ensure correct constructors are called
|
2011-03-25 16:57:28 +00:00 |
|
Frank Dellaert
|
5c193422af
|
Major change, elimination functions are now passed an "Eliminate function", so you can select at run-time which factorization method is used (symbolic, QR, etc...).
|
2011-03-24 19:27:12 +00:00 |
|
Richard Roberts
|
85ed6c306d
|
Renamed typedef name Conditional to ConditionalType
|
2011-03-01 17:16:43 +00:00 |
|
Richard Roberts
|
62b3db9535
|
Documentation for FactorBase, ConditionalBase, IndexFactor, IndexConditional
|
2011-02-15 16:22:35 +00:00 |
|
Richard Roberts
|
12664813d4
|
Avoiding extra work of converting from JacobianFactor to HessianFactor, instead updating HessianFactor directly
|
2011-02-07 22:30:31 +00:00 |
|
Richard Roberts
|
0339a33de0
|
Using Eigen to combine HessianFactors
|
2011-02-07 02:49:58 +00:00 |
|
Richard Roberts
|
bdbc09ba42
|
Added GaussianFactorGraph::sparseJacobian function to create a sparse Jacobian matrix suitable for Matlab
|
2011-02-04 03:56:47 +00:00 |
|
Richard Roberts
|
689adead55
|
Using granular debugging flags, fixed code comment
|
2011-02-04 02:36:14 +00:00 |
|
Richard Roberts
|
24fbe9b02b
|
Correct conversion from unsorted HessianFactor (to sorted JacobianFactor) - will soon re-standardize sorting. Also checking for non-finite values in assertInvariants
|
2011-02-04 01:01:29 +00:00 |
|
Richard Roberts
|
4880257e69
|
Merging hessianfactor branch, Cholesky is now default and GaussianFactor is virtual (see email to frankcvs)
|
2011-01-20 22:22:00 +00:00 |
|
Richard Roberts
|
a8a584d38e
|
Rolling back last commit (undoing merge of hessianfactor branch) - does not compile on Linux, could be compiler or boost version problems. Also will probably break Alex's stuff so will work that out before merging.
|
2010-12-17 01:11:28 +00:00 |
|
Richard Roberts
|
de1892016d
|
Merged in hessianfactor branch, Cholesky is now default. This merge also includes improved timing statements with automatic outlining and low overhead
|
2010-12-17 00:51:51 +00:00 |