[mlpack] branch svn-trunk created (now 5071776)
Barak A. Pearlmutter
barak+git at cs.nuim.ie
Sun Apr 13 21:38:31 UTC 2014
This is an automated email from the git hooks/post-receive script.
bap pushed a change to branch svn-trunk
in repository mlpack.
at 5071776 Modify test for new RangeSearch API.
This branch includes the following new commits:
new 1e06285 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9450 9d5b8971-822b-0410-80eb-d18c1038ef23
new 745f70a Restructure MLPACK directories to the format suggested in #2.
new 895d502 Move some things out of fastlib/ and put them in src/core. I will destroy the FASTLIB name forever!
new 3736684 Finish moving out the last of the fastlib/ directories.
new 2faef60 I have waited so long to do this. I really have, I mean it.
new 68dd90b I have waited so long to do this. I really have, I mean it.
new 7af9c9a Clean up root CMakeLists configuration a little bit, and remove FASTLIB.
new 6f4bf61 Move all the MLPACK methods into src/methods.
new e71f479 Clear up CMake configuration.
new f226414 sparse_censorship was removed so we should take it out of the CMake configuration too.
new b9e896d Update all CMakeLists.txt files to reflect new directory structure. Remove any references to FASTLIB present in those files.
new de2b16e Move everything to src/mlpack and src/contrib. This makes the setup a little more complex, but it makes the build system work properly (although that has not entirely happened yet).
new 592db51 Update all includes so that things work and compile again. A couple MLPACK methods needed to be commented out entirely.
new ceb5079 Add that core.h file which we were using in place of fastlib/fastlib.h.
new f41096a Move header files properly to <builddir>/include/mlpack/ and then install them correctly to $prefix/include/mlpack.
new 990b86e fixed a comment, added paper citation
new be382ab Refactoring math dir : getting rid of unused functions and constants (#116: Investigate what is present and used in fastlib/math/)
new 858959a Remove PI; it is included in math.h, although with less precision.
new 63b6706 range.h and kernel.h are no longer included in math_lib.h.
new 259d2c3 Stop using math::PI; use PI contained in <math.h> instead.
new cc52622 Stop using math::PI
new 6090638 Stop using Sqr(); use pow(x, 2) -- it is faster.
new 6d8ad37 Don't use Sqr(); use pow(x, 2) -- it is faster.
new c808a94 So it's actually M_PI not PI. What am I smoking today?
new 3fb6405 It's actually M_PI not PI.
new 9529042 Include the correct files since math_lib.h no longer does that.
new fad5294 Make sure we include M_PI, and if the system does not define it for us, we'll give it our best shot.
new c67508a Be a bit smarter about what we glob. We don't want .h.swp and we don't want .h.svn-base or anything like that.
new e775429 Make MOG compile. I am certain it doesn't work because I know I broke it in a couple places, but it compiles. In the future, we can do a little bit of cleanup work and it will be good to go.
new 0f9a5b1 Orthogonal range search will be too difficult to preserve and we should start over with it.
new 043f2ac Finish the unit test for HRectBound, completely. Took a couple of methods out and fixed a previously unknown-of error in RangeDistance(arma::vec&). See, this is why unit tests are good!
new 0a51116 Create a simple linear regression class
new b569778 error: ‘struct arma::subview_col<double>’ has no member named ‘ones’ fix
new 6644211 added task sparse coding directory (matlab version) with basic blocks in place (lpsvm and pegasos)
new d4a6edc added first version of TaskCoding optimizer
new 51b8015 updated n_point for reorganized mlpack, also removed link against fastlib in contrib/CMakeLists.txt
new fed0b96 Rename DHrectPeriodicBound to PeriodicHRectBound. Remove the test for now. Also remove a bunch of methods so the API matches HRectBound.
new 0ee4448 trivial change to TaskCoding.m
new bdbf9c7 Change the name of a few methods in PeriodicHRectBound to match. Implement several of the necessary unit tests as a guideline for Vlad, with the most important of those being the MinDistance(point) test.
new feb0786 changed minibatch size for pegasos
new d573052 now passing delta parameter to lpsvm, as it fails miserably otherwise
new eb64650 fixed way of sparsifying codes in generative model
new a0866d4 enabled projection step, although I still do not know if this is necessary
new a5af9c1 added error computation
new 59afe08 various changes to TaskCoding.m
new 1eb197b LinearRegression -> mlpack::linear_regression namespace
new 40571ec formatting changes
new a5967ef fastica -> mlpack::fastica, lin regr style
new 4855331 hmm methods -> mlpack::hmm
new 4c41860 Fix two warnings relating to unsigned/signed comparisons.
new 9ce3790 Don't compile the main executable and test executable into the library itself.
new deebce7 adding (nonworking) single bandwidth test, removed fastlib from contrib's CMakeLists.txt
new c169a89 Update CMake configuration to not comment out everyone's code (only most of it). Remove some artifacts relating to FASTLIB and clean up the file just a bit.
new 727e46c infomax_ica -> mlpack::infomax_ica namespace
new e607db9 naive_bayes -> mlpack::naive_bayes namespace
new 130516d regression -> mlpack::regression namespace
new aae5fec added sparse perturbation versions of Gaussian Sparse Coding
new 7fe1e9e added some run scripts
new 2dda691 Make main usable, --help explains options now
new 24abba4 added code (needs to be debugged) for convex MTFL
new 160cf1b added argyrious convex mtfl
new aac7ee1 Oops, we wanted the name of the file, not the vector.
new 57b62da git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9587 9d5b8971-822b-0410-80eb-d18c1038ef23
new c86dd3f fixed
new ca5c81d Idiot Ryan is idiot idiot idiot idiot
new 7a1d29e mlpack/trunk: modify the clamp function
new 056c823 mlpack/trunk: modify the clamp function
new 6558f3e mlpack/trunk: fix the ClampRange mistake; the previous functions need more intuitive names
new c26d0cb Resolve #142 by removing the problem entirely.
new b1b8210 switched to LP formulation of one-norm SVM, in search of stability
new 4cfc565 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9605 9d5b8971-822b-0410-80eb-d18c1038ef23
new 099d6ce added pegasos minibatch without replacement (bouttou style)
new ad0f36b mlpack/trunk: delete an extra space
new 60bd486 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9610 9d5b8971-822b-0410-80eb-d18c1038ef23
new 7dcec1f git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9611 9d5b8971-822b-0410-80eb-d18c1038ef23
new 38b0149 nnsvm -> mlpack::nnsvm namespace SVMLinearKernel, SVMRBFKernel -> kernels/linear_kernel, kernels/gaussian_rbf_kernel
new 62ff1ab git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9618 9d5b8971-822b-0410-80eb-d18c1038ef23
new dacba39 helps to add the files
new 32fc6e1 mlpack/trunk: per bug 131, add a SaveRestoreModel (and abstract Model) class(es)
new 62044f0 mlpack/trunk: eliminate the newer call to clear breaking the build
new c80a67a git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9623 9d5b8971-822b-0410-80eb-d18c1038ef23
new ea3ad30 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9624 9d5b8971-822b-0410-80eb-d18c1038ef23
new a989213 fixed bug in Learn.m where we accidentally have the same variable name, n_iterations, to represent both n_iterations to run convex multi-task feature learning and pegasos
new 0b2921b git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9626 9d5b8971-822b-0410-80eb-d18c1038ef23
new c233542 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9627 9d5b8971-822b-0410-80eb-d18c1038ef23
new 07db151 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9628 9d5b8971-822b-0410-80eb-d18c1038ef23
new 83ef9a1 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9629 9d5b8971-822b-0410-80eb-d18c1038ef23
new 2cf9c69 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9630 9d5b8971-822b-0410-80eb-d18c1038ef23
new 939e440 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9631 9d5b8971-822b-0410-80eb-d18c1038ef23
new 763fcde git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9632 9d5b8971-822b-0410-80eb-d18c1038ef23
new 251c8be git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9633 9d5b8971-822b-0410-80eb-d18c1038ef23
new 5a924d4 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9634 9d5b8971-822b-0410-80eb-d18c1038ef23
new 7c3f740 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9635 9d5b8971-822b-0410-80eb-d18c1038ef23
new dec3dad changed GenerateSyntheticData so that all atoms are unit norm
new 6cd5522 temporarily changed TaskCodingWStep to use svm_lower_bounded_margin
new eb510d9 added svm with lower bounded margin
new 4ca7243 added pegasos with lower bounded margin
new 5fd0bb5 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9681 9d5b8971-822b-0410-80eb-d18c1038ef23
new 27f5aa0 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9682 9d5b8971-822b-0410-80eb-d18c1038ef23
new 7001d3b changed some display stuff, and now normalize initial W so that each col has unit norm
new c368e7a faster, via precomputation of X * diag(y)
new af6e994 display aesthetics changes
new e64d75b added angle resampling tests, squished some bugs, still a few to go.
new 4a57058 Performance improvements to kd_tree_test, added peer bounds checking.
new 036eb86 Removed some output's that will break the automated unit tests.
new dbdb273 sorry, broke the build; I'm now completely in favor of Ryan's idea to move contrib out of the mlpack directory.
new c61a736 Reduced the size of the kd_tree_test again, that way it will *hopefully* finish in less than 5 min.
new d6e0a2f Clean up code; make helper functions more generalized; grammar Nazi. Use L2 distances because I think we have that implemented just a little bit faster.
new f4314c6 Clean up the code. I think we can make it a little faster by how we do the comparisons. Also there was an invalid memory access issue which is now fixed, in GenerateVectorOfTree().
new e8c6d2e Make lin_reg adhere to the proposed interface for mlpack methods.
new 5a3dcd6 Simple Doxyfile changes. Let's see if we generate the right API.
new 4e04f05 Update Doxygen configuration.
new 359e796 Split IO into IO & Log. Fixed everything up, all tests work. Will proceed to rename IO to CLI.
new 1d29675 For some reason it seems Sed missed gmm and it was already up to date so make didn't touch it.
new 2ba17a7 Shooping the whoop on our Doxygen configuration.
new 363c6a0 Write the main page of Doxygen documentation.
new 7c36047 Doxygenate? Doxygenize? Doxitize? Doxify? Doxenation? DOXY NATION!
new 15308f6 Clean up some more Doxygen.
new 23d4fc3 mlpack/trunk/src/contrib: create my directory
new b86feb3 mlpack/trunk/src/contrib/nslagle: copy the kde code base from D Lee
new 7b4796d added display of sparsity and nnz for Q
new 1ffd8c2 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9907 9d5b8971-822b-0410-80eb-d18c1038ef23
new 7b695f6 added preliminary version of censorship experiment
new 9c128ee mlpack/trunk/src/contrib/nslagle: continue updating kde
new 0c05252 mlpack/trunk/src/contrib: revert changes to the CMakeLists.txt
new a3b31aa mlpack/trunk/src/core/tree: revert the statistic.h for now
new 6d38f9e MaxIP search ported to new MLPACK and ComputeBaseCase is modified to use memmove instead of vector and push_back
new 3d5190a mlpack/trunk/src/contrib/nslagle: commit fledgling kde kcde
new 95515f0 mlpack/trunk/src/contrib: hide PRam directory until the missing file reappears
new 3dcf943 Make a custom stylesheet for our Doxygen output so it matches Trac.
new 8379889 Some changes so it works with Trac. I'm sure I didn't get it all right.
new 5ad3b15 Rank-Approx MaxIP partially completed and not compiled yet
new 24ad621 A couple changes to make it work right with Trac.
new a61a780 Just a couple CSS changes.
new 8e6709d A few more changes because monospace fonts are not being used in the right places.
new 52dac0b Oops, I broke the local documentation style. This should fix it.
new ee0c6dc Add an example kernel for documentation purposes.
new 7c4dc3c Document GaussianRBFKernel just a little bit better and clean up the code.
new bd9d3d3 Apparent inability to perform simple math
new f61e396 Oct 20 arma error
new 03249bf Check .hpp and .cpp files for documentation also.
new db4605a Refactor and comment GaussianRBFKernel. Rename it GaussianKernel.
new 15bdfd1 Comment CosineDistance. Fix a problem I introduced in GaussianKernel. Rename a few files.
new 6aa9777 Document the linear kernel. Rename some files.
new fec7303 Add a few contributors I'd forgotten about and don't include kernel.h -- it doesn't exist.
new 55bc1a0 Update NNSVM. Don't count on those special functions that were in LinearKernel. I probably broke the save/load functionality, but Neil will solve that for us all with LibXML. The future is coming!
new 043f500 Oops, that single warning apparently makes the build unstable...
new 7a926c5 Move files to .hpp and .cpp. Comment LMetric and MahalanobisDistance well.
new d3e2f03 Fix a couple loose ends I didn't before.
new 35a61a7 Fix includes and const.
new c07a630 Update name of lmetric.h to lmetric.hpp.
new d7bef9c Update name of lmetric.h to lmetric.hpp.
new 69ea0fd lots of changes, including loading of bush half censored dataset
new 07b8301 added lars, mostly due to Karl Skoglund and Jingu Kim
new e393f2b added Gaussian Sparse Coding using lars (with simpler code that does not generalize to expfam case, as well as perturbed version, and MTL version with perturbations
new 90f35d0 alt traversal implemented and checked, approx-max-ip still to be checked
new 1bfb327 added negativity constraint to dictionary perturbation
new 01f5f4c added active set algorithms for nonnegative least squares with l1 penalty, as well as nonpositive least squares with l1 penalty (both algorithms are nearly identical)
new 5136a1a added same as last update but using cvx (probably a lot slower
new b0f626c added function for censoring based on some proportion of occurrences of top 10 words in a given topic
new 3eae8cb forgot to add this objective function
new 6262099 added another censorship script
new 64f6cdd git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@9993 9d5b8971-822b-0410-80eb-d18c1038ef23
new ee9c9d4 mlpack/trunk/src/contrib/nslagle: remove my directory from compilation and testing (damn you Jenkins)
new 17c5e4b Document NeighborSearch class.
new d1d0fa9 Put these things inside the gmm namespace.
new c53919b Document the SortPolicy classes; we will use the NearestNeighborSort class as the example class. Also change things so they are .hpp and not .h (#152).
new e6cd2fd Hide parameters inside of a namespace to help avoid collisions (#74).
new d2ce07c Hide PARAM declarations inside a namespace to help avoid collisions (#74).
new 76df992 Approx max ip running, approx tester running as well (but both not thoroughly checked
new 53f5930 Approx max ip running, approx tester running as well (but both not thoroughly checked
new d4ee266 Insane commenting of everything. I think the PARAM_* macros should be split into another file because they are big now. Also, rename some files (#152).
new dc39e76 Implications of renaming io.h to cli.hpp.
new 55272d4 Make sure we include cli.hpp, not io.h.
new 2bbe6f4 Put dballbound into namespace mlpack::bound
new c6f4795 Move log.h to log.hpp, log.cc to log.cpp, and comment those files.
new 153adcd Implications of log.h -> log.hpp change.
new 0e35265 mlpack::math namespace mess
new ea47d3e Hah. That last important file need to change. Yeah, that one...
new 2d2d496 Move some more files to .hpp or .cpp. Place Option and ProgramDoc classes inside the io namespace. Oh, and document them all, because doing that is a Good Thing.
new 3c4a933 Move files and then make sure documentation is created correctly.
new 5ad7972 Uh, I forgot a few things. Stupid Ryan.
new 64c9716 Remove using namespace mlpack::math in favor of math::Range
new 849f636 Comment the rest of the things in the io directory and move the files to .hpp and .cpp.
new c8eef15 Change core/tree, core/math to *.hpp, *.cpp
new 52a8ec4 Use BOOST_REQUIRE_SMALL instead.
new 5707377 change optimizers/* to use hpp and cpp file extensions
new 5b1f44f Change core/model to use *hpp, *cpp Reformat core/model namespaces Add namespace info to nnsmo*.h.
new e38a41b Change linear_regression file names to hpp, cpp Change #ifndef, #define
new 85259de Move nnsvm to *hpp, *cpp files. Fix #ifndef, #defines
new 0ba1b95 Move emst to hpp, cpp extensions Change #ifndef, #defines
new c35eceb Move fastica -> hpp, cpp file extensions Update #ifndef, #defines
new 63fb5c6 Removed Dataset. Converted all data::Load and data::Save to armadillo's .load and .save.
new 2e8c83f mlpack/trunk/src/mlpack: rearrange and restructure the save/restore mechanism; clean up some code
new 6cd2b57 mlpack/src/mlpack/core: remove files I accidentally re-added
new 0813b31 Move spacetree.hpp to binary_space_tree.hpp (better name) and comment it all.
new f6a3758 Update includes of binary_search_tree.hpp.
new b579ba3 Reformat files in the math directory; comment them, and adhere to some new coding standards (#153).
new a350393 mlpack/src/contrib/nslagle: continue modifying the KDE code
new 739e29a Oct 25th, Rank-approximate fixed for very high k-values
new d5931ae Changed a filename
new 38e5831 Changed filenames
new c15d1ea Using math_misc.hpp not math_lib.hpp.
new 60f30ed appropriate changes made to work with new mlpack + alt-angle-pruning added
new 5bffe53 Update names which were missed due to svn merges and my own failure to check the build.
new 10ac3be Waiting for arma::load() transpose -- quick fix now
new 0e346db Bank error in your favor. Collect $100.
new 6e3ee2a Seriously, I didn't check this in?
new 9b875c7 Add extra Mat code to load a transposed matrix. The patch has not yet been submitted to Conrad.
new 89e3e33 Don't include Armadillo directly; just include core.h.
new 142a546 Use CSVs not ARFFs. Transpose the dataset upon loading.
new 9055981 Transpose matrices upon load
new 59ab672 Why was it named .arff when it was a .csv?
new 85bc1ef That's odd, I didn't properly convert that one to CSV.
new da1fe16 Use overloaded mat.save() and mat.load() functions.
new 960b1b4 Different ordering for includes. Include armadillo in arma_extend.h.
new b74c427 Corner case: last_coordinates_ may happen to be initialized to what the user passes in as the coordinates; and in that case, the Precalculate() doesn't actually do its job. So, make sure a precalculation is done if one has never been done before.
new 82d221c Cone-tree code named appropriately
new 69d515a Cosine tree -- co-axial cone tree added and tested
new 98ad632 Require Armadillo 2.0.0 minimum, so we have CSV support.
new 01daf3d contrib/nslagle: newest, not yet working KDE code; multi-bandwidth, dual-tree
new bd29de3 Stupid idiot Ryan should not have checked that in
new 7d20497 Rename files to .hpp, .cpp
new 95c90dd Add load and save and a constructor that loads model from file
new d6660a1 Fix #ifndef, #define, #endif
new b410933 Fix braces, #153
new 3e39bc1 mog/* to hpp & cpp filenames, fix #defines, etc
new 26a6a23 core/file to hpp & cpp
new fe649e4 refactored random generation stuff, wrote a simple timing case
new 5156c67 Stop using TextWriter. This breaks already broken code. Waiting on resolution of #131. Also, this introduces forty ten bajillion warnings.
new b0e02ca Remove col/ and file/ since they are no longer used.
new aa40492 Don't include textfile.h because it doesn't exist.
new a9185aa fixed small bug in angle results, made random generation stuff backwards compatible with boost (hopefully)
new 2c8905c Split timer functionality from CLI to Timers.
new e697c31 Reformatted #includes according to new specifications.
new 150ea32 Fighting with SVN over actually committing changes for timers.
new 063a391 Formatted according to new brace styles.
new c20910f wrote emst_tests, fixed a few problems created by blindly replacing index_t with size_t
new c78b411 Add data::Load() and data::Save() methods to wrap Armadillo functionality. Sadly we can't seem to get away from that.
new cb22fef Re-add data directory.
new 8c0a0b7 Add load.hpp and save.hpp to default includes.
new 4df383c Use data::Load and data::Save. This should fix our tests being slow.
new af8e09b Fixed regression w/ timers.
new 264f400 mlpack/contrib/nslagle: add the multi-tree base function
new 0530e61 mlpack/contrib/nslagle: add the multi-dualtree function
new 629dce9 No need for duplication
new f1d4d70 The all encompasing test, it should be just names mlpack_test in the bit directory. All required dataset files are in a folder as well.
new ef4a472 cleaned up a bit
new 79d59f0 mlpack/contrib/nslagle: commit more changes
new 8281556 mlpack/contrib/nslagle: commit a possibly working version
new ecf550f mlpack/contrib/nslagle: correct for bounds; track the success level for each bandwidth
new e23371c mlpack/contrib/nslagle: commit more tweaks
new 535b4fb mlpack/contrib/nslagle: commit the latest changes; the binary tree apparently is not full
new 1175338 mlpack/contrib/nslagle: I wonder if the algorithm ever really worked...
new 66ef6e3 Use BOOST_REQUIRE_CLOSE instead of just plain assert().
new 9d37d2e Fix compiler warnings so build is stable. This whole output file business still needs to be fixed.
new a1cdd18 mlpack/contrib/nslagle: the algorithm almost works...
new 2ab0f84 Update mog functions, eventually to be renamed GMMs. Work in progress -- not yet complete.
new ecb1578 Add test for MoGEM.
new 944a5ea MaxIP search commit after submission, need to clean up code and add approxMaxIP and fix the dual tree
new 3dc9e8c New pairwise distance computer
new f06f750 Cleaned up and moved all stray tests and test data that is included in the general mlpack_test.
new ae8f64b Someone owes me some food for this... well... if it's the bug.
new df7ae59 mlpack/contrib/nslagle: KDE almost works, again...
new d1218f9 comments
new f78debd vec = double resized the vector to size 1.
new e3b5632 A better test, but it's not quite working for some reason.
new 31c5234 Set intercept to 0 for test debugging.
new 0268c78 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10206 9d5b8971-822b-0410-80eb-d18c1038ef23
new 71aac2d Remove the extra load() stuff that was not the right choice in the end.
new f872e33 No more contrib/.
new 258eefc Fix documentation a little bit. We want the bug list; we don't want .svn.
new d2f1799 Use new load().
new 94d4d34 Linear regression changes & fixes
new 03af748 Better comments, coding style improvements
new 560314b Ticket #126 : PCA code
new 6887b37 Ticket #126 : added entry for pca
new 06cea80 Update all tests to new formatting standards and make them all .cpp not .cc.
new 45c2f20 added comments and made some formatting changes
new 0ffe721 Add HMM test. A bunch of stuff does work.
new d40b1be Modify HMM code. We'll make a new class, HMM<Distribution>, which will serve the purpose of all three of the previous classes.
new 00f9de4 Change the names of a few methods, and implement supervised training/estimation.
new 1b952ee Add test for supervised transition and emission matrix estimation.
new 61fed12 Loosen bounds on test because the variance in the test is a little more than I had thought. I think these bounds are still good enough to know the test works.
new d7ee57f Now there's a tricky corner case. If the random value is 1, then that inequality is never satisfied.
new 862fd6e Implement Generate() functionality.
new 744f484 Widen error tolerance to more reasonable levels (determined empirically).
new 3dd6245 Add function to calculate log-likelihood.
new f57c3c5 It's tired and I'm late, so I make stupid simple errors like this.
new c6e7a9d Add one more test for Generate(). The tolerance is a bit higher than I would like, but that's okay.
new d4ae82e Test HMM::LogLikelihood() for the discrete case.
new e3c8c21 Comment Estimate() a little better and provide a convenience overload.
new 46887e7 Use exact tolerance for probabilities, because it's harder to estimate smaller numbers.
new 9aa8932 Begin construction of Distribution classes so that we can generalize the HMM class.
new 2725303 Oops, I named a file wrong.
new 927199a Update CMakeLists to include new files.
new fb3f5cc Flesh out DiscreteDistribution class entirely. Inline functions in class definition.
new 64e51b8 Add tests for DiscreteDistribution. These all work.
new dbbd866 Fix compiler warning.
new 33bead1 Make normalization safer; avoid division by 0.
new 9e5ce55 Add simple constructor and then make sure that every constructor normalizes the distribution.
new 3469f96 Make sure the distribution is normalized.
new 03de9df Test that the distribution is normalized after construction.
new 706421d Abstract away from the discrete case. Now, everything depends on Distribution (I think).
new 7e32ab4 Use new HMM syntax, with DiscreteDistribution instead of int.
new 5f3fdf6 Document the HMM class more effectively.
new 3e5c098 Use a default template parameter for HMMs.
new 9907152 changed the pca api. not using Armadillo's pca anymore. Still making a few changes ..need to check-in the comments to the api.
new eaa5269 Code refactoring. Added comments to the methods.
new 0bb7a8b Move mog to gmm and update CMake configuration.
new 8a45eaa Remove MoGL2E. That can be implemented some other day.
new fef9123 Move GMM test to where all the other tests are.
new 916057c Add an overload of phi() which can do multiple data points at once; this way, we can avoid recalculating the inverses of the covariances over and over and over again.
new 676781b Remove debugging output.
new 84ee5ca Rename MoGEM to GMM. Make cond_prob calculation a lot faster (but there are still some speed issues to work on).
new 17c76d9 Why did this ever pass any tests? I am officially confused.
new 6776365 dd
new a083ef3 Checkin mid-progress. Arbitrary data types added, need to clean up NearestNeighbors since it makes a few (too many) assumptions about the matrix it'll be using.
new c5c30c0 What is wrong w/ my svn?
new ca51507 Working copy SVN metadata is broken..
new 52e943a Diffed a clean working copy, this should end my svn issues.
new 6bd3f95 Reverted changes, as it is apparent diff did not work. Will work on that, too.
new 887a594 Format things just a little better.
new efddf2c I am really frustrated by tests that don't always have a deterministic outcome.
new 9c539a2 Update GMM code. It should be a little faster training, but it is still too slow for my preferences. I am not sure what is making it so slow.
new d52ab35 Compile GMM.
new 3460858 Flesh out main executable for GMMs.
new 7871cc8 Tests for GMM. Multivariate multi-Gaussian training tests included (that was the hardest one to get working right).
new c6f3bdf Const-correctness, yo.
new 1182a25 Fix some warnings I inadvertently introduced.
new c9cb59a Add Gaussian distribution class.
new 77a3ce9 Compile Gaussian distribution class.
new 558fd59 Tests for GaussianDistribution class.
new 1002fdb Why is this file here?
new 83ec0fe #162: use log-likelihoods instead of direct probabilities because direct probabilities get too small.
new b135f87 Add a first test for a Gaussian HMM.
new 1428a3a I accidentally had not generalized to the general observation case.
new 09cff1c Add test for Gaussian HMMs where we are training for both the labeled and unlabeled case. Almost done with the Gaussian HMM tests.
new d0a9761 Widen tolerances for unlabeled training.
new 31e746a Fixed some formatting.
new 30bd390 Test Generate() with a Gaussian HMM.
new f9972e8 Make sure the state probability makes sense.
new c330b49 Don't use srand() in tests, so that the results are reproducible.
new 63667a8 Move distance functions and metrics into metrics/, and update the build accordingly.
new daedc53 Update kernels to metrics (where applicable).
new 4c13567 Update kernels to metrics (where applicable).
new 82978e8 Fix style as per #153 and name local class members as per #118.
new dcce751 Update style for #153; it was not completely done.
new 02e9792 Adapt style per #153.
new 18be42f Fix formatting as per #153.
new bbf5ced Remove test file, and fix formatting as per #153.
new 3bb5686 Update style as per #153. Also, we didn't need the test here anymore.
new ae9c8ba Fix formatting in tests/ as per #153.
new 7f8a1cc Format EMST as per #153.
new 81cc122 Comment FastICA code as per #153. Remove a couple unused functions in lin_alg.hpp.
new 1f87f2e Change filenames as per #152 and fix formatting as per #153.
new af0ff38 infomax_ica.h is now infomax_ica.hpp.
new 40394b8 Remove kernel PCA before Ajinkya's version, since Nick's won't be used.
new 55173a6 No more kernel PCA = no more kernel PCA test.
new 008ecfd Update formatting of linear_regression to reflect decisions made in #153.
new 50f228c Move filenames to .hpp and .cpp to help finish #152, and fix style in accordance with #153. Removed phi() since it is a copy of the version found in methods/gmm/ (which will also have to be moved).
new d221e8c simple_nbc.h is now simple_nbc.hpp.
new abb7d5d Change NCA filenames according to #152 and change style according to #153.
new e3b168f nca.h is now nca.hpp.
new 224b269 Rename files to .cpp and .hpp in accordance with #152, and fix style to be in line with decisions in #153.
new bcb8695 neighbor_search.h is now neighbor_search.hpp.
new ccd05d0 We won't use Nick's implementation of ridge regression; we'll extend linear_regression instead.
new 9ac7fed Remove RidgeRegression test since we aren't using that implementation anymore.
new 06b1261 Change to .hpp.
new 036dd11 Change name to .hpp.
new 1871d99 arma_extend.h is now arma_extend.hpp.
new 2429cd9 No more regression.
new 639507d arma_extend.h is now arma_extend.hpp.
new 923d45c Use long doubles for comparison (compatibility with older Boost UTF).
new 25585cd Style MVU as per #153 and move to .hpp and .cpp as in #152.
new 8734051 Fix formatting of NNSVM code (#153).
new ed40af2 These methods are unmaintainable and will have to be re-implemented.
new bf76fba Remove SVM test and NNSVM test.
new 0133383 This is not used anymore.
new bc8c02b Move core.h to core.hpp. Fixes #152.
new acf2961 Move K-Means to be its own method.
new c16989e Make KMeans into an actual class. It now supports "overclustering".
new 003362a Add test for KMeans.
new 2537817 Add KMeans test and comment out GMM test since its API has not yet been adapted to work with the new KMeans.
new 791e148 Don't build GMMs because they don't yet work with the new KMeans class.
new c73be0c Add test for MinDistance identical case.
new 13213a0 Add a test for the identical case for MaxDistance().
new 4515594 Adapt to new KMeans API.
new bd2fd77 Start compiling GMMs again.
new 94c9cc7 GMM tests work again.
new 827812b <#126: Implement simple PCA> adding pca_test
new 976b84b <#126: Implement simple PCA> adding pca_test
new 01551c2 <#126: Implement simple PCA> removing pca_test from pca dir (Added it in tests dir)
new b89374a git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10418 9d5b8971-822b-0410-80eb-d18c1038ef23
new 37373be <#126: Implement simple PCA> changed assert to BOOST_REQUIRE_SMALL
new bc5a92b Absolutely minor formatting issue fixes; use BOOST_REQUIRE_CLOSE instead of BOOST_REQUIRE_SMALL.
new f5a7b9e Abstract some features of KMeans into template policy classes.
new be7eee9 Adapt KMeans to new policy-based API.
new 0547185 Test the different policy features of KMeans.
new fe4f97e I thought I was helping, but it turns out I just made things worse. BOOST_REQUIRE_SMALL is necessary because the eigenvalues could be zero. The fabs() isn't, though.
new 322f7ba You have to zero the counts first!
new 44a05b5 A better way of calculating the covariance, more akin to arma::cov.
new d8b572e Executable for K-Means. Celebrate!
new cf69c2e Main executable for PCA. A couple little tweaks to the PCA code here and there.
new 24ce665 <#165: PCA::Apply() fails with Armadillo error> was a minor bug in the indices passed to shed_cols
new af87108 A couple formatting fixes after Ajinkya's fix for #165.
new 8804395 A couple more formatting fixes.
new c6f4f5c Make sure the covariance matrix is positive semidefinite.
new 6bd7330 Add another test case and simplify the first test case; the second test case currently does not work.
new 20ee989 Update timers.hpp and timers.cpp to conform to formatting guidelines (and fix a syntax error!).
new 25e6c8d mlpack/core: apply some changes to HRect, Log, and BinarySpaceTree
new ce70714 mlpack/core: undo an unnecessary change; remove a pesky warning
new 3abed91 mlpack/core: add std::vector as a save/load valid type (so long as the templated type is basic, for now)
new 100177d Change API to newer standards for HRectBound and BinarySpaceTree. Fix some formatting issues introduced in r10459.
new dfa399a Make Range adhere to new formatting and naming conventions.
new 42d52c0 Update uses of BinarySpaceTree to conform to new API.
new a9499e3 Update tests to reflect API changes.
new a9f1b11 I have truly gone full style Nazi. But hey, now the variable naming conventions are correct here!
new 206217a Improper use of regex leads to over-confidence about code compilability then sudden erosion of programmer confidence. Moral: your regex is always wrong.
new 2f9dd81 Revamp L-BFGS code to remove CLI from its insides. Naming convention fixes, const fixes, the works.
new 9571306 Fix const-correctness of GetInitialPoint().
new ee04616 Remove FastICA because Nishant will be implementing RADICAL instead.
new 0336cdd Don't compile lin_alg_test while I think of where to put that code.
new 9d7d41a Fix for integer parameter.
new 0fda9e1 Move FastICA linear algebra utilities to core/math/ (pending rename).
new 160cfe5 Run the linear algebra test again since it's now in a different place.
new ce3b874 < #166: PCA fails comparison with MATLAB> centering and scaling was causing the difference. By default other packages do not do zero-centering and scaling with std deviation.
new f70f05b Remove unused PARAM_*.
new f6bda85 Ripped CLI out of nbc & gmm. Moved over to their respective mains.
new aa53e04 Ripped optionshierarchy out of CLI, deleted files.
new b436f65 Overhaul API for NeighborSearch (it probably is not complete) and remove dependence on CLI to help fix #150.
new e219849 Update tests to conform to new NeighborSearch API.
new 5f18ab5 Accidentally committed this minor change.
new 3852bc2 Remove CLI parameters from LBFGS.
new 74cae8c Remove infomax_ica since it will not be released.
new 5ca86a8 Update comment to be more accurate.
new 71895b5 Minor tabbing fix.
new 7d4fa0c Remove InfomaxICA test.
new e8d9778 Remove Print() function in accordance with #164.
new 1e99b50 Formatting fix, and prune CLI from the internals of DTB.
new 7a1dc64 Prune CLI from the internals of DTB.
new e87c4c6 Put the statistic class into the tree namespace.
new e07328c Add (as-of-yet) untested Range::Contains(const Range&).
new aaafddf Add RangeSearch to the list of methods we have.
new 31ad18f Add tests for RangeSearch; they are comprehensive (pretty much).
new 4f96351 Add Dimensionality() function to distributions; use arma::mat and arma::vec instead of std::vector<Observation> and Observation; and don't use size_t in DiscreteDistribution.
new 83231a9 Adapt test to new distribution API.
new 6734784 Adapt HMM to only use arma::vec as observations.
new f6b43be Adapt HMM tests to new API.
new 1d74562 Compile RangeSearch.
new a158753 Fix sanity check to actually check things correctly.
new 1c7fb94 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10528 9d5b8971-822b-0410-80eb-d18c1038ef23
new d9026a9 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10529 9d5b8971-822b-0410-80eb-d18c1038ef23
new d0736c7 Add Estimate(const arma::mat&, const arma::vec&) so that this class can work as a distribution for HMMs.
new 9e41629 First simple test for GMM::Estimate(const arma::mat&, const arma::vec&).
new 9fa2c74 Move .h to .hpp and .cc to .cpp. We don't need dummy.cc anymore so it's removed.
new 8f417c5 Compile LARS.
new 9f86aca Big API changes for LARS. Move things into the constructors; rename internal variables. Did a little commenting but I can't comment all that much.
new 3c62649 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10543 9d5b8971-822b-0410-80eb-d18c1038ef23
new 23dbdb7 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10544 9d5b8971-822b-0410-80eb-d18c1038ef23
new a513b22 changed CMakeLists.txt for radical so that compilation works in .../mlpack
new 4fc6cef added radical to CMakeLists.txt
new e409d1c Apparently an adversary swapped every occurrence that should be data.n_rows to data.n_cols and vice versa. This has been fixed, and LARS works once again
new 6823979 added LARS tests, but still need to make small changes before adding to CMake
new a88a3cd added test cases for lars
new 56cda74 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10567 9d5b8971-822b-0410-80eb-d18c1038ef23
new c7c06b2 Add GMM test case for training multivariate Gaussians in the setting where probabilities of each observation are given.
new b7c75e6 It helps when you actually implement the stupid function instead of being an idiot and thinking "well I'll just do it later but actually I'll forget that I didn't do it and then not get to go to sleep instead".
new 98ebba7 Issue a warning when an empty GMM is created.
new b58e442 Test GMM::Random().
new e188460 Actually, make that debugging only...
new 1446182 Test HMMs with GMMs; a simple Predict() test. Still need Estimate(), Estimate(...), LogLikelihood(), and maybe some other tests. I am reasonably confident the whole thing works correctly.
new 4dabd73 changed the test to accomodate pricipal axes in opp direction ie when eigenvectors returned are flipped in sign compared to the eignvec returned by matlab for the same dataset
new 85d474c changed the api to accomodate centering and scaling of initial dataset. There was a bug as the end result of the transformed data was not centered, the results were not matching those from matlab.
new 125d45f changed the api to accomodate centering and scaling of initial dataset.
new 4d3e694 Don't use srand() in a test case.
new 0f7bdfa Don't use srand() in test cases.
new f2cb9b8 split dtb into _impl.hpp, should finish off ticket 117
new 459c0e0 Add -NDEBUG flag.
new c2d6a5a Transpose before adding rows.
new 217173c Misspelling of option.
new 583dbb3 various style fixes to radical
new 6e257a7 more style fixes
new 6c6efa7 I think I fixed two bugs. need to check tomorrow
new aab4479 Reimplemented CLI w/o heirarchy. Eliminated concept of modules from CLI and main functions. So if it was foo/bar before, it will now be just bar.
new cfaf90d Did some more formatting of output. Added more shortcuts to methods.
new 4707273 we no longer store the data matrix matX as a class member, we changed the call to WhitenX (it is now called WhitenFeatureMajorMatrix), and added comments
new 7a42171 put Radical in namespace mlpack::radical and moved WhitenFeatureMajorMatrix out of Radical class but into mlpack::radical namespace
new 29b3a65 added lots of comments
new 526d877 small fixes to radical_main since we present radical with a n_dims by n_points matrix
new 76104ee made some speed optimizations via manual profiling
new ef62703 some speed optimizations
new a9e9dcc For some reason, this caused a segfault with -O2. This needs to be looked into.
new 82e68ca added data needed for Radical test cases
new 49ed26a git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10685 9d5b8971-822b-0410-80eb-d18c1038ef23
new bc05a38 updated radical main
new 3f49bca added test case for radical
new dc60708 added radical_test.cpp to CMakeLists.txt for test cases
new 88a6c21 updated radical_test.cpp to use data::Load rather than .load
new 4abbea8 fixed small filepath bug
new 28a818c you can now set the input and output filenames, as well as the options, as paramters
new 917f2fd cleaned up lars_main and added command line options
new db3fda0 stylized lars
new f78917e removed unnecessary files
new 99dd441 removed unnecessary file
new 6ec0c60 fixed CMakeLists.txt for lars and switched from _impl.hpp to .cpp
new f30e0db git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10699 9d5b8971-822b-0410-80eb-d18c1038ef23
new b4c7e23 many style fixes, made some functions private, commented out some unused functions
new e3a765c cleaned up lars
new 4bea638 and i present you <Kernel PCA>
new e7a2633 Yeah, break everyone's build by requiring 2.4.0. It's for a good cause though.
new 61f064c Add sparse support which can be patched into Armadillo if necessary.
new ffc22c2 Convert u32 to uword for 2.4.0.
new aa47e74 Templatize the matrix type.
new 23674c5 Abstract-ize HRectBound to accept arbitrary types of things.
new 090ce88 A typedef so other types that depend on the matrix type can use it.
new 0e22be5 We can use regular .col() here since we abstracted away arma::vec.
new d370af2 got the implementation into hpp file to include it in the class def. This is need to make the templates work. Good article here which talk about why we need this http://www.comeaucomputing.com/techtalk/templates/#whylinkerror
new f90c30d git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10718 9d5b8971-822b-0410-80eb-d18c1038ef23
new 7a2ad4c Stop using unsafe_col().
new e2c0535 Refactor API to be consistent with RangeSearch.
new aa39d01 API changes.
new 211bb11 LARS no longer has X and y as class members. They are passed into the main computation function DoLars
new cf6a2a7 updated lars_test to reflect changes to lars
new fd6cd17 These aren't necessary with Armadillo >= 2.4.0.
new a892015 Too much noise means I start doing things wrong.
new fcdd855 I had been approaching the wrong solution to the issue because they keep bringing me drinks.
new b851861 Copy code over from arma-sparse branch.
new 4a83915 Retry the API change. This time it works...
new 217f60e Simpler swap_cols() and SpValProxy::operator=(SpValProxy&) to make it work.
new 234c18d Why does this fix the test? I don't know, but there's not complete time to fix it right now.
new 8823555 Allow sparse vectors to be passed to metrics (this should get a bit of speedup). Now we leave the computation up to Armadillo.
new 2ed63a9 Abstractize vector type.
new be5ddd7 Abstractize k-means so it can use sparse matrices.
new dc7cd19 Explicit conversion to mat.
new 5b7ec43 Centroids should be sparse, too, if we want them to be.
new 9394ff9 Support SpSubview proxies.
new 832367f Use a const reference internally in SpSubviews.
new ebf959f Test sparse K-Means.
new 3d652c5 Unused variables.
new f7369db Abstract-ize matrix type used for NCA datasets. The covariance matrix should not be sparse, though.
new 5350662 increased tolerance for RADICAL test
new 2272703 Sparse NCA doesn't actually make sense.
new 5044884 removed random seed from radical_test.cpp
new ad1c878 Abstract-ize NBC to allow sparse matrices.
new 6b0f1f7 Clean up CLI formatting... for like the fifth time.
new 112c761 Document EMST better.
new 1c6b352 Oops, I need a <>.
new d84583b Make statistics useful again.
new 9ce5c13 New Statistic API.
new e3aedf0 New Statistic API.
new 56cee90 Restructure test.
new 912cffc Refactor and clean up EMST code.
new 80aa0fb Refactoring of EMST/UnionFind API.
new b3186f1 Update for transposition of output.
new 4268804 Clean up CLI output again and again. Then change how Timers works because it was being destructed before CLI in some cases. That's the danger of singletons...
new babc5d8 Make it how it was before.
new 020e5fe Allow compilation with -DARMA_EXTRA_DEBUG.
new 1aa24e2 Compile libmlpack.so not libmlpack.a.
new 22cb95f Comment things a little better, and rename one of the timers.
new 6a95a0a This was not an input parameter.
new fe46576 Change in HasParam() changes a couple things in this test.
new a844d83 Move io/* to utilities/* then utilities to util (shorter, simpler name) and update all references thereto.
new 719a58b -v == --verbose
new 03ad900 Allow a GMM to be saved and clean up documentation.
new 5db963a Minor documentation fix for the namespace emst.
new eeeb694 Documentation fixes for GMMs and removal of now-unnecessary README.
new bdcd8cc \bandwidth is not valid LaTeX... thanks regex...
new a3493d7 Two asterisks, not three. What is the meaning of life?
new ec5d407 Last run of modifications for EMST documentation.
new 55e170f Expand the example for the GMM. This documentation should be done.
new 26ef5de All these files are unnecessary because I re-implemented HMMs.
new 770925a Update documentation for HMMs. Documentation should be done for this class.
new a011a0b Don't build these targets.
new 5fc40c6 Remove these executables because they don't work anymore.
new 60bf3d5 Better documentation for K-Means.
new 459939d Did some refactoring on CLI, fixed a potential bug. Also removed some references to PARAM_MODULE.
new 1b50e44 added some comments to LARS
new 6e1e7bf more comments for lars
new fb4cb40 Fix #174 and another bug in the main executable.
new 9c05c6c penultimate set of lars comments
new 4374f4b Fix segfault.
new 01158cc moved accessor definitions for LARS from lars.cpp to lars.hpp
new 1788166 Copy test data to the right place after building the test.
new 6ebabf0 Standard header include protection.
new 69e9d9e Split math_misc.hpp as per #143.
new 6d0ab1c Change variable names to match code specifications and mark a couple functions const.
new faf688d Add license and copyright to satisfy #175.
new 3fcea27 That's embarrassing, I forgot the name of the guy who opened the ticket to make me fill this whole thing out...
new e42bb92 Ignore some other things that we hadn't previously been ignoring.
new def79f5 Move kpca to kernel_pca; clean up the code to match formatting standards (okay, some of my changes were superfluous), use size_t instead of int where necessary, and change CMake configuration so kernel_pca is compiled.
new 1236f48 Clarify program name.
new cddb207 Redo LinearRegression API and put it into the regression namespace (not linear_regression). Make the main executable more functional and document it.
new a551bc6 Namespace and API change for LinearRegression.
new 3c204f6 Fix a bug which was overwriting memory and may be causing that segfault I am trying to hunt down on the build server...
new bbd6fd0 Refactor and comment NaiveBayes a little better.
new b0bf6d5 Fix bug and better warnings.
new 3ecff99 Document NCA.
new a690e07 Not sure why I need to raise the tolerance... maybe something worse is wrong?
new e43d01d Redo API for NeighborSearch for the last time and go over documentation.
new 35b717f Change formatting of RADICAL code and redo the documentation a little bit to make it more consistent with other programs.
new e5e8a5e This caused a problem for release-mode, so maybe this is the instability in the test.
new a041714 Reset the tolerance to normal.
new b659b52 Format LARS a little better and give good program documentation.
new e468d86 Update LARS test.
new 0a21d50 No longer need l here.
new 327e130 I can't reproduce this error anywhere! Is something wrong on the build server? Comment out the test for now.
new 4b395d3 Clean up warnings.
new ed5d0e8 Version .so at 0.99.0.
new 8da5862 Remove unnecessary 'normalizer' variable and add accessor for gamma. No mutator because those two variables are codependent.
new bcfba80 Use accessors and mutators for math::Range.
new c6bc473 Update tests to use accessors and mutators for math::Range.
new 057a848 Add accessors and mutators for AugLagrangian and L-BFGS. Change int to size_t where necessary.
new 294e0eb Refactor BallBound API and clean it up.
new 288d9f6 Change in BallBound API.
new b7cda97 Some SpMat bugfixes.
new 1faf0c0 Document timers and CLI a little better since we changed how it all works.
new 7055b06 This isn't used anymore...
new 00a096b Add some useful documentation.
new 66205cc Better documentation. Now we have a quick guide to MLPACK.
new 15d04c6 Remove a few warnings.
new ff7e656 Pretty poor implementations of min() and max() for sparse matrices.
new fd15860 Clean up BallBound code; now, expand things in batch because that's better.
new 5d2f93f Clean up warning, use Info instead of Debug.
new 6721a0b Clean up MVU code and add a main executable to it.
new a087f97 Make sure that Doxygen actually picks up the guide we wrote.
new 1bb8bfb Inline math::Range.
new 0952521 Use Boost random number generators; account for API changes in 1.47.0.
new 1194863 Require boost::random and link against it.
new 5bf7136 Stop using rand() and use Boost random numbers.
new 55eee01 Explicitly specify namespaces.
new e2308e9 Give a few specializations for vectors, and remove using directives.
new 2cd01f5 Give direct access to discrete distribution probability vector.
new 5627a29 Update to new direct-access DiscreteDistribution::Probabilities() API.
new f0be566 Clarify error messages, and actually assign the parameter to what is read for a few cases. Use std::string& not std::string for LoadParameter().
new 7aa474b Formatting for the entire /tests.
new d641be3 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10856 9d5b8971-822b-0410-80eb-d18c1038ef23
new dbbd748 Turning in with some tests for the implemented distances in periodichrectbound.
new 9867e89 Properly formatted /gmm
new d396a5c More test cases, as well as formatting of /kernel_pca
new b2bbb87 Correction for the range test.
new db9ede1 Formatting of /kmeans.
new d25872d The test cases for maximum periodic distance, as well as the code for both versions of the maximum periodic distance.
new 11bf545 Removed the extra print statements that I forgot in there.
new 8698a9f Formatting of /linear_regression
new 17f0a47 Formatted /mvu
new 03fd5e4 Formatting of /radical
new 6fb4f02 Formatted /pca
new fed9217 Formatted /nearest_neighbor
new cb318dc The fix to the previous fix.
new fcb18db Formatted /math
new ea6f630 Formatted /util
new 260c430 Formatted /tree
new e88f827 Normalize vectors because Probabilities() doesn't do it anymore.
new b6347b0 Add main executables for HMMs.
new 132c7a9 Explicit namespace usage.
new 56e605c We have a few more executables now.
new 43cfd4d Implement operator|= correctly.
new 033fef5 Fix use of uninitialized variables.
new 0cac5de Utilities for loading/saving HMMs; to be deprecated later.
new 0ceaeaf Move distributions to core/dists/.
new a50f9cb Add CMake configuration.
new 8e46f1c Fix formatting for L-BFGS.
new 6728ea8 Fix formatting for AugLagrangian and associated test functions.
new 14b6d74 Fix braces.
new 8008dc8 Fix comment.
new 348b9d3 Change name of box().
new 8967121 Boost <= 1.41 did not have a separate random package.
new 862d003 Search correctly for Boost.Random. It did not exist separately until 1.46.
new e7227d5 box() -> Box().
new d1b3fd5 Fix >= Boost 1.47 issues.
new 716c7a6 box() -> Box()
new 3d28b48 Smarter linking of MLPACK against Boost.
new 4e519d5 Fixes for old Boost.UnitTestFramework.
new 26207ce kernel pca works with linear Kernel now.
new ef3346d git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@10910 9d5b8971-822b-0410-80eb-d18c1038ef23
new 4107e42 update version and soname
new 2bae162 Don't run test cases that don't work because the code was never finished.
new be5a02a Fix two warnings and some style.
new 78a72d6 kernel pca test file
new 628e61c exponential kernel
new ec7e107 laplacian kernel
new 818f8c1 polynomial kernel
new f658349 fixed compilation error for boost 1.47 by adding <> to uniform_01
new f84d0cd formatting changes
new 692e5a1 hyperbolic tangent kernel
new f9dce7c fixed a few bugs
new 270bf94 tests for the new kernels
new 9747eb4 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@11090 9d5b8971-822b-0410-80eb-d18c1038ef23
new 76f9831 Fix all the things cppcheck didn't like.
new 278cddd We don't actually need a reference to the new child...
new a2ead52 non-class type return values are non-modifiable, const qualifier is ignored (-Wignored-qualifiers)
new f808e64 Comment out unused parameter names (suprisingly valid C++), -Wunused-parameter
new 9d39c6e How many unused parameters does one library need...
new 060d4f1 Add RandomSeed() function to set the random seed, which is always a good time to be had.
new 92db4a8 Use uint32_t because that's what the class uses internally.
new ccfeff0 Fix the compilation error from my -Wunused-parameter fix when compiling without debugging and profiling and with -O2.
new 7c9cc0b Another -Wunused-parameter fix
new 900a52b Add -Wextra and -Werror for debug and non-debug builds.
new 9794dcc Apparently jenkins uses an older version of boost and we had a #define that caused an unused variable warning...
new 10e8f6b Hey, cool! A warning solved a bug.
new 09459e9 WTF SVN? I update, you give a conflict, and when I resolve it, you drop the semicolon? Or was it my fault...?
new b716114 Add option for random seed (for #182).
new f8fdcd0 Add support for --seed to HMM executables (where necessary).
new 8503bdc Add --seed support for K-Means executable.
new 23aac69 Minor spelling fix.
new cb60880 Set the standard seed (srand()) also, because Armadillo is likely to use that, and in places we use .randn() or .randu().
new 85fcfea Weird, these aren't necessary.
new 7533cc7 Make documentation more the same-like. You know?
new d3bc5ea Add a tutorial for NeighborSearch. That took a little while...
new b2b873c Add an actual page for the tutorials.
new 056da81 Some formatting fixes. We have lines longer than 80 characters but that is unfortunately necessary to make Doxygen work right in some cases...
new 7018b61 So that we get that nice hyperlink.
new 56f76b9 No -Werror for now (the build server gets angry, gcc doesn't have to).
new be0b0e9 Fix some errors present on older gcc versions.
new ca2f1f2 fixed some stupid typecasting warnings
new 8663756 No need to link against LAPACK or BLAS; Armadillo does that.
new e5be655 Install binaries in the proper location.
new 2881b63 Wait, it's not an ugly hack anymore (it was before, then I changed it and forgot to remove the comment).
new 3543936 Add 'make test' command. This should make the build fail if the test does not pass. It is not run by default if you type 'make'; you have to type 'make test' to actually run the test.
new a0c9ff3 Add 'make doc' command, but it does not yet install the documentation (that will come next).
new a86ecc2 Rename project to 'mlpack' (so we install documentation to /usr/share/doc/mlpack/ not /usr/share/doc/MLPACK/) and add installation rule for generated HTML documentation (goes into /usr/share/doc/mlpack/html/).
new d25f097 Higher tolerance for Laplacian kernel.
new c318ff7 Don't use GNUInstallDirs. Assemble /usr/share/doc/mlpack/ by hand (it's not so hard!).
new e20f81a Don't install Doxygen documentation if it was never built.
new 921ed02 The parameter names were... wrong.
new 7c9fb63 Haha! Other people should be able to make tutorials too!
new cf92982 Fix scoping issues. Stupid Ryan!
new 9857ca5 Refactor AugLagrangian for better use by other classes. AugLagrangianFunction is its own standalone class now to allow for easier specialization. The tests aren't quite done yet (the LRSDP Lovasz-Theta ones still fail and give unnecessary debugging output) but that's the next thing to do.
new 8ac948a This will hold an optimizer for low-rank semidefinite programs, formulated by Monteiro and Burer. It was originally using AugLagrangian but I decided to specialize it further to the general LRSDP case.
new 82730aa Basic implementation of LRSDP. Not yet tested.
new cd45103 Don't set any RPATH.
new 6238b51 Revert RPATH change. We have to build with the RPATH and strip it at install time.
new 7d85d69 Actually write the "correct" installation commands, so that RPATH gets stripped at install time.
new 28eb99b Add operators for SpSubview.
new f3670dd A script that can turn our output into man pages. This is for packaging, but man pages are good, you know.
new aa27993 No preview in an automated setting...
new 0c1c0d6 Script to make all man pages.
new d551e2e Make the directory if it doesn't exist.
new c67f597 Force gzip compression; make sure it is not interactive.
new 25b5475 Make rules to generate man page documentation. This is goodliness.
new d63dc50 Apparently, when I wrote this, I did not understand how floating point numbers worked.
new b8d97d6 Temporary work check-in so I can work somewhere else.
new 89c6471 Temporary work check-in so I can work from elsewhere.
new 9050cf6 Change documentation a little bit to prevent man macro error (hopefully).
new 0878fae Fix spelling error.
new 995b71e Wow, awk is crazy. This fixes warnings in the troff output given by txt2man.
new de13cce A few changes to the implementation of LRSDP. Still does not quite work yet.
new eb7dc1e Limit on number of iterations, and set the first lambda value right for LRSDP (this will need to be changed later).
new 5e3c784 It works! I can't believe it works.
new 19f3690 Don't do extremely simple LRSDP test; it needs a different first Lagrange multiplier. We'll work that out later. Remove SDP tests from AugLagrangian tests. Compile it all and add it to the usual tests.
new 22075e4 Allow setting of random seed for RADICAL.
new 7b2c25f Fix a few warnings for unused variables that I introduced.
new 5be7674 Add the dataset we work with.
new 7585c96 Allow two modes of constraint matrices. Not yet documented.
new 3328879 Add tests for two other Lovasz-Theta problems. hamming6-4 isn't converging right, and keller4 is working, but takes 5 minutes. Perhaps johnson8-4-4 will be the only test case we use in the long run.
new 25ff40b Implemented range search, need to implement output functionality. Also need to update the documentation to reflect realities of range search.
new 8d720fe Default and copy constructors.
new 166583b This fixes the arm test bug.
new 0c8cc41 Update for new L-BFGS API.
new 9f5b85f Revamp L-BFGS API, moving around where the maximum number of iterations is set and stored.
new e65a9da Revamp the AugLagrangian API to deal with the new L-BFGS API, and take a pre-constructed L-BFGS object at construction time.
new c08265b A quick hack to make the tests faster. This'll go away when I redo the LRSDP API.
new 0395ab2 Revamp NCA and MVU, which use AugLagrangian and LRSDP, respectively. MVU is not yet done but, bit by bit...
new a51c92e Don't link range_search_main.cpp into MLPACK; it adds unwanted CLI options. And use spaces, not tabs.
new 1533445 Further refactor AugLagrangian API to make it simpler.
new 3224f78 Refactor LRSDP API for ease of usage.
new fb4a50f Clean up LRSDP test and add a dataset which we are not using (but the test is written for it -- it's just commented out).
new 92a78fd Just so it compiles for now...
new c93eced Let's not set sigma to a value which ensures the algorithm will never terminate...
new 26984ae Fix out-of-order initialization warning.
new 4b1169f Various improvements to LinearRegression and our executable.
new 5928bf3 Incomplete rough draft of the LinearRegression tutorial.
new 34706f7 coding with LinearRegression examples
new 5a743e1 more comments for the comment god
new b312ba3 fix ticket #184
new b78ef76 Add RandNormal() and fix some const-correctness semi-issues.
new 487f966 Misplaced space. The horror!
new bac6310 Initialize StatisticType objects in the root node too.
new a705472 MRKDStatistic for mrkd-trees. Probably still needs improvements.
new d0209f9 A few fixes so that things can compile with older Boost.Test versions.
new 5aa7319 Some changes; use M_PI instead of math::pi() (for consistency across all MLPACK code), and explicitly specify using namespace mlpack::math so that radical compiles on older versions of gcc.
new f5b6452 Make this actually compile when used with BinarySpaceTree
new ac25d39 Appears to work now.
new 5670ec8 Add a splitDimension variable and getter for determining which dimension a node is split on.
new 261447f Tab fix (think I got them all)
new 95ceb3a fabs(max(,)) -> max(fabs(),fabs())
new d0cf188 Move to range_search_main.cpp and clean up executable.
new b16b662 A couple changes to the kernel PCA test.
new 7672722 Clean up some comments and the sanity check on the dimensionality should be the upper limit, not the lower limit -- dimension < 0 is not possible with size_t.
new 2998efe Put down the crack pipe, Ryan.
new 727ce35 This is a copy of the Gaussian kernel... oops...
new 100f240 Remove exponential kernel from sources.
new 7839e54 Some cleanup for kernels.
new c59141a No more ExponentialKernel.
new b40c1f3 An actual main executable and some implementation tweaks for speed.
new b9da0f0 corrected elastic net / lasso objective function in the comments. I had forgotten to put the 0.5 scaling factor on the reconstruction error part of the objective
new f3c6c9f Some formatting fixes.
new f07b0aa Add --scale and --nocenter options.
new 0930f38 I had the options backwards; also clarify documentation.
new 15d095b Add --scale and --nocenter options. Also center by default for PCA.
new 2ff6966 Helpful note pointing to other documentation, so we don't need to include citations in -h output.
new 31f0bcd Slight change because CLI outputs differently now.
new dd99bae Remove unnecessary citations and change formatting of citations with respect to opinions expressed in #201.
new e5afc55 Remove citations from -h output.
new ec8890b Remove citation from -h output.
new 5245688 Remove citation from -h and put @code around it in the RADICAL documentation.
new 18ba9b5 Fix references in tutorial; rewrite into a little bit and clarify some terminology here and there. Also unnecessary grammar Nazi strikes again...
new e2340ef Forgot two references.
new be678fa Fix references.
new 3cd65c6 List linear regression tutorial.
new 8323efb Credit where it's due. I might be missing a name or two...
new 576a43a Mistype found.
new e1e22ac Add a range search tutorial. Should be done...
new 7b54e12 Add range search tutorial to list of tutorials.
new 5ab59d6 Removed c-strings in favor of std strings.
new 9b87535 That isn't a correct Doxygen comment...
new 4bd2c35 mlpack/core/kernels: add a couple of kernels appropriate for the forthcoming KCDE/KDE combo, per #200
new dbdb7f0 Initialize LRSDP A and b vectors to the correct size at construction time, instead of confusingly expecting users to take care of that themselves.
new 7e9ff38 The sparse A matrices should be able to have a value.
new 66723d1 Changes according to the changes in the LRSDP API.
new cd8318e Hey, so, it turns out when you find the package, you've gotta add the include directories too.
new 323c749 These aren't necessary any more.
new 065f676 Some changes to mvu. Not 100% sure if it works yet.
new d169ce1 added sparse coding
new ca04361 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@11671 9d5b8971-822b-0410-80eb-d18c1038ef23
new 53cf489 added small 500-point version of MNIST
new b75b449 fixed CMakeLists.txt to reflect correct demo program name
new d5313ed fixed small bug in lars: sometimes after kicking a dimension out of the active set, we also need to explicitly set the corresponding coefficient to zero. In that case, we were saving beta to the end of the beta path before the fix. now, we save beta after the fix. This fixes a bug
new ff16e76 fixed a bug in how we were checking the optimality conditions for the LASSO. holy shit, an error in the test case itself!
new 8190f7e added test cases for sparse coding
new 82cfd73 updated path to mnist file
new 60b2694 added modifier (setter) for data X
new 8e0cacb added sparse coding to CMakeLists.txt
new fa59909 added sparse_coding_test to CMakeLists.txt
new 1577332 added local coordinate coding to methods
new 6a7a19e git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@11683 9d5b8971-822b-0410-80eb-d18c1038ef23
new 310693b added local_coordinate_coding to CMakeLists.txt
new 075ae14 Mostly, hopefully, working implementation of Pelleg and Moore's mrkd tree k-means.
new 82330a8 added tests for lcc
new 42d731c Some changes that had been forgotten.
new 2ae47c1 Unpack tar.bz2 for Nishant's sparse coding tests.
new 47765fe Maybe it works, but I'm not totally sure.
new 8b928bc Modified string types in cli & related sources.
new 3381aa3 Three new methods for trunk.
new 931b569 Add local_coordinate_coding and sparse_coding to man rule.
new 3424bd4 removed some commented code that no longer is needed
new 196db0c Untested Windows code needs to not be untested...
new d2d87eb Fix #207.
new 2142c2a when useCholesky=true, we now use a once-computed Gram matrix for inner product computations between points. If one is calling LARS repeatedly with the same data X (as in sparse coding), one should use SetGram even when useCholesky=true, and in particular, it is better to use SetGramMem
new 901b85e Switched to Cholesky-based LARS. this version of LARS now uses Gram matrix in some places. We use SetGramMem rather than SetGram to avoid copying Gram matrix unnecessarily. Also, removed some commented out that was quite old
new 9e9e705 Install range_search executable.
new eb22cd4 Something's wrong with this
new 4ef4e08 Use in-house random functions, not drand48().
new 05ed096 Make CMake configuration better by linking directly to the library locations, not just the library. This is for Windows compatibility.
new d7053c5 Reformat CMake configuration file. Only use -Wall and -Wextra on GCC-like compilers. The special handling of LibXml2 is unnecessary because that's all done by FindLibXml2.cmake anyway.
new 27490de We shouldn't provide these ourselves; CMake 2.8 does that already.
new c2b9a07 Windows #defines min and max; undo these defines so we can use std::min() and std::max().
new d8f9f2b More typenames, required by VS2010 but not by GCC (for some reason).
new eafaae1 Only give stacktraces on Linux boxes; Windows doesn't have execinfo.h or cxxabi.h.
new 0512910 Fix apparently untested Windows code.
new c94c02d No execinfo.h on Windows.
new b1414f1 Eliminate ambiguity in a few function calls.
new b2c1586 Require FORTRAN because FindLAPACK requires FORTRAN.
new d6ec5e8 Wait, we don't even need LAPACK. That comes through Armadillo.
new dc7174d Turns out not every version of CMake ships with this...
new 04c40c9 Special handling of Boost linking properties for Visual Studio. This was necessary for me to make things link correctly.
new 4bdeb3f Implementation of round() for C99-deficient compilers (MSVC...).
new 9c0c18d Use boost tgamma and include that round function we made earlier.
new f3cdd4c Link to direct location of unit test framework library.
new a263798 Various fixes for Windows compilation.
new 7a192d0 Be a little pickier with how we build on Windows. Static library because creating a DLL correctly would be a nightmare.
new 9f6b930 Fix #210. Stupid error...
new 3554811 Generate Doxygen documentation with 'make doc' only in out-of-source build directory -- don't clutter source directory. (I'm proud of this)
new 448f101 Make timers multi-run.
new cadd9f5 Test multi-run timer functionality.
new 3b8cb65 Look for possible dependencies of libxml2.dll, to issue a warning to the user in case they don't exist.
new 9bbac7d Link against LAPACK and BLAS on Windows because Armadillo doesn't automatically do it.
new 7d28f49 No colors on Windows because command.exe doesn't recognize ANSI color escape sequences...
new 381eb9a Patch by andreasl: no definitions of promote_type<u64, ...> are necessary if ARMA_64BIT_WORD is defined.
new 7c30dcd Patch by andreasl: change u32 to uword (the generalized Armadillo type, which can handle 64-bit Armadillo).
new 0c6437e Working, mostly finished pelleg and moore mrkd kmeans implementation
new f51702d Add more fields to the mrkd statistic
new 4f3f720 Oops, I did screw it up.
new 0acec12 I think that some of the promotions I had written are actually invalid. For now, we simply use what is given in Armadillo when ARMA_64BIT_WORD is defined.
new 9ac6374 It makes no difference in reality, but don't define is_supported_elem_type<u64> or is_supported_elem_type<s64> if ARMA_64BIT_WORD is used.
new 6f226aa comment fix
new fc67a05 Define timersub() before it's used on Windows boxes. For some reason VS2010 didn't have a problem with that.
new 4345352 Use Sleep() on Win32 systems. This isn't tested yet -- but I'll do that in a moment.
new 9114f75 Greater than or equal to -- because some systems are very precise.
new 7781131 Require version of libxml2 to be at least 2.6.0 (which is dated late 2003... so it shouldn't be a problem for people to have it).
new 89d3694 Search for iconv.h on Windows systems.
new 41bccf0 Artifact from pthreads which isn't used anymore.
new 1b73a14 Add GMM::Probability(arma::vec, size_t) as part of #212.
new 24f8783 Test GMM::Probability() (both overloads).
new 8d4efd8 Add a Classify() method.
new 0db3917 Test the new Classify() method.
new ceb6eaf Increase tolerance a little bit so the builds don't fail so much.
new cff5d67 I hate test tolerances so much.
new d53bf84 Forgot to commit this, fixes forgetting to set minDistance
new 4bacdfd Added enhancements to pca & kernel pca.
new 3faccef Undo accidental test compilation commit -- we like all the tests.
new e50917a Add to contributors
new a664130 Update metrics to support distinct vector types as arguments.
new 55a66be remove useless code
new e0a7d09 Modified PCA/Kernel PCA interfaces to remove redundant centering option.
new cec3429 Density Estimation Trees (DET) added to mlpack
new 2af45ab DET CMakeLists added
new 1aaab15 DET CMakeLists
new 9c6d050 DET naming schemes
new 527568b DET naming schemes
new b3e0160 Handle a case where the step size ends up being 0 but the gradient is not yet the minimum gradient size. Maybe it is a little slow (elementwise comparison is not incredibly fast)...
new 8659afc Clean up some man page warnings.
new 61f3e5c DET Tests (almost all done) added
new ae90d3d DET tutorial 1st draft
new 98c4d50 Use Log::Debug instead of cout for debugging output (maybe it should be Info?). Also needlessly replace '\n' with std::endl -- really, there's no reason to do this; I just did it.
new 16dc612 Declare the function before we use it.
new 8974a75 Refactor GMMs so that you can use any clustering method you like. The default is "EMFit", which is the EM algorithm (exactly how it performed before).
new 6c28c5d Use GMM<> not GMM, because it's now a templated class.
new 00ec665 Rewrite GMM tests to use GMM<> not GMM (because it is a template class now).
new 66b84f7 Switch from armadillo's load() to data::Load()
new c2de4cf Fix tabs
new f9e390d Fix test compile time errors, missing <> for GMM
new cceb82b The fabled GMM-HMM training test. The hardest of all tests, only the strongest training implementations survive the double-sucker-punch of GMM followed by GMM. A more rigorous drill sergeant might require tighter tolerances, but unfortunately perfection can't be expected out of the EM algorithm every time.
new 02800c7 Hackery necessary for sparse support using Armadillo 3.0.0 or greater.
new fba46e9 Include new traits.hpp file.
new 0adf805 Workaround for Armadillo 3.0.0 bug.
new fccafc3 Fix ambiguity error under Armadillo 3.0.0.
new 9e1ee74 Pass one: comment grammar Nazi and fix code style issues.
new bd566ce Pass two: const-correctness (mostly) and use size_t instead of arma::uword.
new 53f9a5c Missed a couple uwords.
new 6900184 Rename some variables in the class. Not done yet...
new 7f04e42 Update tests to reflect new method names.
new f629023 Fix longer-than-80-character lines.
new 6322ae3 Fix longer-than-80-character line.
new 9025a76 First pass: style. Fix a memory leak or two too.
new 8dfb529 Fix namespace ambiguity.
new 4ea75ed Take care of FittingType when we are passing in means, covariances, and weights to build a GMM.
new b9cc366 Handle Boost 1.33 Random library correctly. I am not sure how far backwards this fix extends.
new 2ab535f Fix preprocessor ifs for Boost <= 1.39 Random API changes.
new 9792687 Make ourselves compatible with older versions of Boost.Test.
new 6a08db6 We need to actually get the version number before we check what it is...
new d4e6b87 Zero the memory before we start adding to it.
new 8ebd074 An implementation of cover trees. It is not yet polished, and the debugging comments are still inside of it (though commented out). I will be polishing it shortly...
new 7ccb14c Add two tests for the CoverTree class; a simple test and a big (random) test.
new 428bf0e Add NumChildren() and Child() to BinarySpaceTree for compatibility.
new 029d53a Fix warnings I introduced. Oops...
new 1ac6ffd Comment the CoverTree implementation a lot better. Still a few things to do...
new 6fe40c6 Remove unnecessary debugging output from test.
new d918b28 Regresion is cool, but this method is regression.
new 5b5696f Refactor LARS. Fix comments, remove unnecessary methods, and consolidate constructors.
new 111e453 Don't use a pointer for LARS objects; no need to do memory allocation on our own. Adapt to new constructor calls, so there is no more need for SetGramMatrix().
new 2a231a3 In fact, remove the line entirely.
new 482beab Refactor in accordance with new LARS API.
new 56dd32c Allow arbitrary metrics to be used in the construction of the cover tree.
new ba479fd Test an alternate metric construction of the cover tree.
new 2e2b48a Fix misleading documentation...
new 9971559 s/IO/CLI/g wasn't a good idea.
new 001fbc9 Allow template parameter class to specify which point should be chosen as the root node. Document template parameters a little bit (but hopefully this will be redone at some point).
new fe40258 Add NumPoints() and Point() to BinarySpaceTree. Maybe Count() and Begin() could be deprecated...?
new 697c3fb Give better debugging output.
new 5efc0c8 Tabs to spaces, and return a reference, not a pointer.
new 94ba25d Add a single-tree depth-first traverser. It is not as fast as it could be.
new 40105c8 Add the NeighborSearchRules class, which defines how the SingleTreeDepthFirstTraverser can perform a NeighborSearch. Adapt the NeighborSearch class to use this. It is not as fast as it could be.
new 43889a8 Ack, how could I violate my own coding style guidelines?
new 14abb49 Add the very similar SingleTreeBreadthFirstTraverser, meant to be used with the CoverTree.
new ba71c8b Modify the API of BinarySpaceTree to include a few more methods. Starting to settle on a more standard tree API.
new 27f5d3f Modify CoverTree API towards a more standard API. It can be used with NeighborSearch now... although it probably will fail miserably.
new 73f7218 Update our traversers so that they don't consider self-children. I am not too happy with how unclean it is, but I expect it'll get cleaned later.
new 7dc77b5 Update API to use TreeType::MinDistance() because other tree types may not have a bound object.
new 2b0d523 Use the tree's preferred traverser type.
new e7f773f Fix some formatting and style issues, and then add a test for nearest-neighbors cover tree.
new 5e84561 Remove unnecessary debug output.
new afb103f Add a dual-tree depth-first traverser.
new dea44bb Save just a little time where we can.
new 350bc8c Include statistic.hpp (so we have EmptyStatistic).
new 0dec288 Add dual_tree_depth_first_traverser.hpp to build list.
new e38e95b Add an option to use cover trees.
new b0b5ffd Add implementation of rules for dual-tree search.
new 5768f85 Use DualTreeDepthFirstTraverser for naive and dual-tree calculations, and remove all of the guts we previously needed but don't anymore (ComputeBaseCase, ComputeDualNeighborsRecursion, ComputeSingleNeighborsRecursion).
new a3b6a9a This bothers me a lot more than it should. (used to say "5mins,4.3secs", now says "5mins, 4.3secs")
new d11123f Make this an iterative implementation.
new be2a999 Well, that would break everything...
new d0da579 Add DualTreeBreadthFirstTraverser (at this moment, not rigorously tested).
new 0729345 I didn't do this right; I forgot about the "UpdateAfterRecursion" step and broke everything. So... revert to the recursive implementation for now.
new e039828 Correctly handle cases where implicit nodes are created.
new 528930f Add a preferred dual tree traverser.
new 216d858 Add preferred traverser and dual cover tree traverser (which doesn't prune at the moment).
new 307d136 Add rules for cover tree search. This breaks the nice abstractions I was trying for.
new f8fe1a3 Add a dual-tree test for cover trees.
new 2ab8799 Add directory for implementation of MaxIP. I think it needs a better name...
new f98b46f Remove debugging output.
new 068f0e9 Add the IPMetric, which is the "inner product" metric which calculates the metric induced by a kernel.
new ba9dee6 Add rules for the MaxIP single-tree search.
new bdc1e12 Add MaxIP class.
new cfc2c56 Add CMake configuration so MaxIP gets compiled.
new 879df6d Add maxip to project build.
new 3c4f258 Add simple first test for MaxIP.
new fed5f1e Clarify error message; k can be equal to the number of points.
new 8597262 It helps if the conditional is right, too.
new 7a64408 Give number of prunes.
new fb251ec Add maxip executable.
new 41ae684 Don't get it backwards...
new d3ecd82 Add template specalization for linear kernel.
new 1c10810 Avoid double-evaluation of base case.
new 40bc6e2 unsafe_col() is faster.
new 935f7fd UL and ULL extensions for older GCC.
new 7d30d4e Reimplement single-tree search. It's faster now.
new f39babc This file doesn't exist. I didn't need it.
new 9b03b46 Ensmallen and randomize.
new c17dfb5 Initialize products to DBL_MIN.
new b55d371 Use unsafe_col() 'cause it's faster.
new 84f8716 Rewrite to only use one queue. Not very happy with this implementation.
new 1acd75d Add a kernel which I'll remove later. I don't like it...
new 211f8e7 We can do polynomial kernels too.
new 7428670 Check in modifications.
new c8989e9 Update kernels in ugly ways that need to be removed later.
new ce7c92b Needs to be the square root, not actually K(q, q).
new 557233b Put ugly global variable in so it compiles. This will be cleaned later...
new 47d9361 Consider queryProducts kernel evaluations too.
new f821de4 Don't used squared distance in tree construction.
new 9ef1907 Revamp breadth-first descent to consider tree levels so it is the proper breadth-first descent.
new 23286e5 Count kernel evaluations.
new c17f6d0 This test does not compile with Armadillo 3.2; I have not yet had a chance to figure out why.
new bf3ee9f Work for older CMake versions (2.8.2) and finding older Armadillo versions.
new d2bd281 Halfway transitioned to newer implementation, which does not try so hard to conserve memory.
new e8281a8 Rewrite swapping code.
new 8ce0e43 Some slightly ugly additions -- but which give an order of magnitude speedup during cover tree construction. It turned out that when we were handing the new near and far set to a child, we did not remove points with distance further than the far set bound, meaning that the far sets were way larger than they needed to be. Preliminary results indicate fewer distance calculations during tree-building than Langford's original implementation.
new 41d083b Relax the separation constraint to include equality (in accordance with the paper).
new 842aac6 Store furthest child distance and furthest descendant distance. This can allow better pruning rules.
new d23d671 Fix a couple bugs in furthest child and descendant determination.
new ec0f020 No far set in root node.
new debfbc6 No far set in root node.
new 10c2350 Parent distance is the interesting one to track, not furthest child distance. Fix a few bugs so that furthest descendant distance is correct.
new c7cef2c Add two more kernels. They need to be parameterized.
new b758608 fixed line that referred to nearest neighbor search
new 02702f0 wrote EMST tutorial
new c91b38d Clean up cover tree implementation a little bit. Allow instantiated metrics (good for hyptan or gaussian kernels or any other kernels with parameters).
new cb7de62 Refactor and clean epanechnikov kernel.
new 280e032 Remove static Gaussian kernel hack.
new af13bb7 Best not to promise functionality we might never get around to.
new 81a27bd Remove the no-offset kernel; it is not useful and was a hack for a deadline.
new e5e75b7 Include new files.
new c95643e Fix top-of-file comments.
new a902256 Rewrite triangular kernel with parameters.
new 5e6078f Check in mess of MaxIP code which is being cleaned little by little...
new 0ca5f9d Test modified for new MaxIP API.
new 2ef3c4b I don't believe this is a valid kernel.
new b2fd2a1 Update documentation slightly.
new 1048c01 Finish refactoring for each kernel type.
new 7e79051 A couple of files were not included.
new e6061e5 Some code cleanup.
new b0db6ec Code cleanup; don't use namespace arma (since that's done nowhere else) and clean and condense things a little bit. Not a big overhaul -- but it's coming.
new 7ec040a This was being passed in backwards.
new 9372070 Modify LARS API to accept non-transposed matrices. This is easier on the user...
new 8437362 Adapt for new LARS DoLARS() API.
new cddf749 More refactoring; remove nActive because it isn't necessary, and include solution in the call to DoLARS().
new 9935965 Changes to DoLARS() propagating to here.
new 8c2681e Changes to DoLARS() propagating to here.
new 44baf0f Change DoLARS() to more descriptive and less awkward Regress(). Minor code cleanups, and turn comments into full sentences (grammar Nazi mode activated).
new 89bcd19 DoLARS() -> Regress().
new 13eea02 DoLARS() -> Regress().
new 590b754 Apparent typo in the test cases meant one combination was not being tested.
new 435cbaf Changes to main executable in accordance with the modified API.
new f1f83ba Add option allowing user to specify if they want the data transposed on input; this is useful for LARS.
new 206e571 Tests for new options to data::Load() and data::Save().
new aa585e3 Clean up LARS executable.
new 20345b7 This check changes because we're loading X and not transposing.
new e1f76ec CLI::GetParam<bool> seems to crash, and, add a timer.
new 88429c0 Add some timers for loading and saving operations.
new 6dcd981 Clean up documentation slightly.
new 73aa22b Stop timer for early exit.
new 940832e Clean up comments for no good reason.
new c638825 Refactor SparseCoding to initialize the dictionary as a template parameter.
new d02c28f Change tests to respect new SparseCoding template parameters.
new fa6f705 RandInt() and Random() sped up by just using randomUniformDist
new e8976ce A couple of formatting fixes.
new 391ade5 A couple more tutorials now exist.
new 5e25cc3 Added a compilable version of NMF code. Test cases need to be written
new 11befda Basic implementation of NMF
new 6decd09 Add nmf directory
new e7d0301 Move everything into the NMF directory from PCA
new 6e2ed99 Move NMF out of PCA
new ecbb2c2 Revert to PCA CMake configuration
new df58d6f Clean up comments, and store a reference to the data, instead of copying it.
new 7ff1f15 We don't need to set the reference -- this line is now implicit.
new 0f220dd Rename DoSparseCoding() to more flowing Encode().
new f40f645 Unit test for NMF added successfully. Validation conditions need to be tuned
new 0900173 Minor style changes.
new 45a4751 Fix some syntax errors I introduced, and print the matrices. The test does not appear to be working correctly...?
new 1d3626d Formatting fix, and handle things a little more carefully.
new 760e47d Clean up code. Use std::pow(x, 2.0) not x * x. A couple of formatting fixes, and some changes to make some code more efficient.
new a750a50 Stop using activeAtoms because it isn't necessary (minor speed boost). Clean up code just a little more.
new c8656cf Avoid a matrix copy for matActiveZ and potentially speed up gradient calculation by avoiding instantiation of arma::ones<arma::vec> (Armadillo may have already been smart enough to avoid that anyway).
new 981dd02 Clean up main executable. Avoid unnecessary copies.
new 70b61e8 Fail on failure to load dictionary.
new fcf053b Avoid unnecessary initialization of dictionary when the user passes a dictionary already.
new a60aea5 Clean up style violations.
new 6430dfa Clean style violations in main executable.
new 701f598 Template LocalCoordinateCoding to accept different kinds of dictionary initialization. The loading dictionaries from file feature is temporarily broken for the LCC executable, but this will be fixed soon.
new cf65cfc Update for new LCC API, and fix style violations.
new e1e2273 Make OBJ_TOL and NEWTON_TOL parameters to the methods.
new 12121ef Change matX to data, matD to dictionary, and matZ to codes.
new 8543c88 Update for new API with better matrix names.
new 7363b7c Clean up formatting of output, and stop using arma::uword (prefer size_t instead).
new 54212a7 Oops, this isn't necessary anymore.
new c37a64f More consistent comparisons
new 177d29b Change DoLCC() to Encode() and some cleanups and fixes in Encode(). Reduce number of redundant class variables.
new 9f448d1 Fix inaccurate comment.
new 5585013 Update test cases for different cosine distance.
new f7308f1 Comment out testing procedure for now because it doesn't work (yet).
new e9bbaf8 Clean up and optimize code.
new 1abbb98 Move RemoveRows() to the linear algebra utility section since it's used in more than one place.
new 30c8ba8 Include linear algebra utilities.
new a12658e Move RemoveRows() out of class definitions.
new 036c103 Remove extraneous spaces
new 46bf742 Minor spacing and formatting changes.
new fc16a11 Cleaner calls to .rows() as opposed to .submat() with all those arma::span objects.
new b6517e1 Tests for RemoveRows().
new eb97555 Be pedantic about namespaces.
new 5ce21f0 This causes an ambiguity error (mlpack::math or arma::math?) on gcc 4.4.x.
new 969e7ab Set limit on maximum iterations so we don't spin forever.
new f71de1d Better error messages (more descriptive, less awkward).
new 76a2555 Severe cleanup of executable, so it actually works again.
new ad934b4 Use timers internally, not externally.
new 55fcd17 Use timers internally, not externally.
new 89eb220 Fix unclosed '
new 6465f68 Fix doxygen documentation (forgot a @code).
new ad04089 Use log negative error to prevent calculation overflows (a first step towards making cT and eT unnecessary). The algorithm has not yet been changed... so LogNegativeError is always used as '-std::exp(LogNegativeError(.))' at the moment.
new 4749629 Update test API to use LogNegativeError().
new 2f2743e Simplification of expression
new 3d3679f Yeah, this is simple enough to inline; probably a better choice.
new cf630c8 This function can be const.
new b8e227f This function can be const.
new 8e1ab7c Added Aleternating Least Square method to NMF. Require some testing
new 6c43eed First shot at overhauling FindSplit_(). Break deep nesting of if/for by reversing if conditions. Use negative log error instead of actual error to help combat overflow issues. Use double instead of cT (long double). Change default eT to double.
new 9d2783b Fix lots of formatting issues in DETTest tests. Stop casting to/from float/long double to preserve accuracy of calculations (which are now done on doubles).
new d8a9835 Minor style fixes and const-correctness fixes.
new 2b24781 Uh, I forgot the opening {
new beb1c20 Further cleanup of casting issues.
new 69f8789 Apparently unused file
new b3d67c6 Add parameters so user can specify objective and Newton tolerances.
new 5eb47aa Make objective tolerance parameterizable.
new 465fd4d Switch to doubles not floats in main executable so it compiles.
new b1f49ac added some commented out debug statements
new 7e8bd49 Clean up casting for old Boost.Test versions which aren't very smart.
new ddeed90 added back in accessor for matUtriCholFactor
new 06b6566 Rename FindSplit and its variables to names that adhere to the style guidelines. Remove a few unused variables.
new 94a0718 Update test to modified API.
new e7c3a0e Fix corner case where found version is equal to the required version.
new 4a7fce2 Add PSpectrumStringKernel to be used for maxip/FMKS.
new 52bd91d Add tests for PSpectrumStringKernel.
new b89f3cb Clean up FindSplit() and reduce API to only what is necessary.
new 83ccc5b Clean up test to adapt to new FindSplit() and SplitData() APIs.
new 5e31a23 Replace tabs with spaces. Remove unused variables. Use separate BOOST_REQUIRE_EQUAL() lines instead of one big BOOST_REQUIRE() with a bunch of &&s in it (this can give better output if one of those is failing, whereas the version with && is not specific as to exactly what failed).
new 9b734eb Clean up constructors. Don't hand-allocate memory for minVals and maxVals; remove GetMinMaxVals_() because it was only ever used once; inline it directly into the constructor where it was used.
new 2432b23 Clean up test. Don't use 'new'/'delete'. Fix wrong index for TestSplitData.
new 8bcd50b Oops, there was one more 'new'/'delete'.
new d0099c9 Clean up constructors; use initialization lists more exclusively. Also clean up WithinRange() (inline it among other things; it is short).
new 3410808 Use BOOST_REQUIRE_EQUAL(), and the function is now called WithinRange().
new b811118 Further switching from cT to double. Later we will switch to all log error.
new be69849 Use double instead of long double for Grow() and PruneAndUpdate().
new 5f2a9a5 Use double for ComputeValue(), and take a reference instead of a pointer; also fix const-correctness of function.
new 325e0dd Update for changed ComputeValue() API.
new 5ef66b0 Change API for ComputeVariableImportance().
new c497a42 Modified API for ComputeVariableImportance().
new 86a454e Stop using eT and cT for good. Change all variable names to adhere to naming guidelines.
new afc8a7a Refactor test to stop using eT and cT. Don't typedef MatType anymore, and change variable name to line up with naming guidelines.
new d22d37b Clean up main executable file.
new d662360 Further cleanup of variable names.
new 54a5c5c modified OptimizeDictionary to return norm of gradient of Lagrange dual (with respect to dual variables)
new be289c0 fixed test for sparse coding dictionary step - the dictionary step now runs for more than one iteration, and the check is to see that the norm of the gradient of the lagrange dual (WRT the dual vars) is sufficiently small
new cb8a8ca Convert all of DTree (with the exception of ComputeVariableImportance()) to work in the log domain. This was fairly difficult...
new 708e579 Adapt tests to working in the log-space and update APIs accordingly (as they have changed).
new 14e2136 Cleanup of dt_utils.hpp. Get rid of 'using namespace std' in dtree.hpp and update accordingly.
new 1a5af14 Remove templatization of density estimation trees (it wasn't necessary). It will be brought back, but not at this time.
new 832a1f6 ...and make sure that the removal of templates is propagated to the test file...
new fb8b0cf Clean up CLI parameter names and then rename the executable to 'det' not 'dt_main' so it matches more with other MLPACK executables.
new d541f39 Fix std::pow ambiguity on older compilers
new 772e5b5 Fix pow ambiguity that I missed last time...
new 2d69d77 Actually, I got it backwards, I need pow(double, int).
new 718e02b I added in logic to remove duplicate flag parameters in response to Ticket #231. http://trac.research.cc.gatech.edu/fastlab/ticket/231
new 30f496b This day is off to a great start
new c9bf2ed Remove unnecessary debugging output
new 9688bc3 Fix DET so it actually works. A few things here and there needed to be changed.
new 501058f Disable volume regularization for the main executable for now.
new 777874e Remove debug output which is unnecessary now.
new 5c52fca Rename various parts of NMF to different names which line up more with the style guide; clean up main executable documentation and allow the choice between different update rules.
new cf9f3ad The API call for Apply() has changed slightly.
new 8ad153c Allow objective tolerance option.
new 9dfc2ff Fix tabs into spaces.
new d176668 Wrote all the NMF test case and updated Alternating Least Squares method with pseudo-inverse function'
new e744b9b Restructure trees entirely. Each tree will have its own SingleTreeTraverser and DualTreeTraverser, and each tree type should be kept in its own directory (for organizational purposes). The previous traversers in traversers/ have been adapted to either the BinarySpaceTree or CoverTree and moved there.
new b055864 Add a few files we forgot, including the convenience include files which include both trees and traversers.
new 16ece03 Change include files and APIs to use the new tree traverser setup.
new 248f9b5 Modify tests for new include locations.
new a1cf67a Minor cleanup of NMF code; I think the residue should be displayed in non-debugging mode (optionally with -v) so I switched to Log::Info. Comment on the change to pinv() then rewrite RandomAcolInitialization a little bit to avoid allocating memory unnecessarily. Unfortunately insert_cols() is a little slow because it allocates more memory and memcpy()s.
new 71f156d Add Score() and Rescore(), which will be used for a "new" type of tree traversal which should be faster.
new c5503ba Refactor BinarySpaceTree traversal to use Score() and Rescore() and thereby avoid unnecessary repeated calculations (so I hope).
new 69c530f Remove two now unnecessary files.
new 856d43a Oops, those files had to be removed from the CMake configuration too.
new dab90bb Oops, we need this check in there too.
new 56223e4 Use unsafe_col() instead of col() for speed reasons (this makes a BIG difference).
new 220169f Ensure we check if we can prune before descending the query tree (this case is possible and it seems, happens quite frequently).
new 914f402 Fix comment.
new f8bc89c Give ourselves a way to force inlining.
new 3af9aa5 Check if we can prune individual points before performing computations.
new fbeaf51 Speedups for math::Range().
new a4db591 Inline these functions for speed.
new 2d9c5ff Force inlining of BaseCase() because that is extremely important to loop unrolling optimizations. I mean like 200-300% speedup important. Huge!
new f72534a Respect -s when combined with -c (single-mode cover trees).
new ff46bbd Update single tree cover tree traverser to be smarter about recursion order.
new 4fa249a Add new methods for finding the maximum or minimum distance to a point or node if the distance between the point and node center (or node center and node center) is already known.
new 5660796 Add new methods for calculating the best distances when the distance between the center and the point (or center of the other node) is already known.
new 22c5f7a Remove a few unused methods (LeftFirst() variants). Add Score() which accepts a pre-calculated base case, which is useful for trees with self-children, like the cover tree.
new abe9831 Rewrite the single tree traverser for the cover tree. It works, now it needs to be sped up.
new a317d81 Fix spaces, and actually traverse to the left instead of just saying it and then doing something else.
new c4b94dc Fixed NMF Test
new e51fc1c RandomAcolInitialization was modified to take a template parameter so now we must specify the <>.
new 82f51ad Heh... a template parameter is not an lvalue. So just warn the user and then chug happily away...
new 94bd10a Oops, a small typo...
new 0e980af Tighten bounds on minimum and maximum distances because we already cached the furthest distance.
new eaad119 Revamp single tree traverser... again. For now it has debugging output.
new b9ee0a0 Use L2 distance not squared-L2 distance because that breaks cover trees.
new 72c2979 Get the conditional right. Only map things back like that if there was a query file and we were in single mode. Not just 'singleMode' is good enough.
new 1b6b07c Refactor single tree traverser again; don't use priority_queue which is slow. Instead exploit some nice properties which allow us to use a map of vectors. It seems as though this implementation is about twice as fast.
new 980ecde FindArmadillo.cmake ships with CMake from version 2.8.5-rc1 onwards. So require that.
new 18c6fc6 #223: I don't think we have ever needed to link Boost.Math anyway. Also use -O3 for shits and giggles. I didn't see much runtime difference but whatever.
new e6c9e8a I promise, GCC, just this once, I'm smarter than you, really.
new 0c600e4 Split CoverTreeMapEntry into its own file since it will also be used by the dual tree traverser.
new a794052 Some warning cleanup.
new d9d87c3 Seeming error; we just want the path, not the absolute filename.
new c6925e9 Use correct L2 distance, not squared L2 distance in cover tree test.
new 6c66e6b Rewrite dual-tree cover tree traverser. The use of UpdateAfterRecursion() is kludgey but it seems to work okay for now. Also, the MapEntryType couldn't be abstracted across both traversers because the DualTreeTraverser needs more information.
new e4bc775 Remove unused CanPrune().
new a0860f3 Be explicit and call std::log(double) because in some compilers (icc) std::log(size_t) doesn't exist. #239
new aa3f3c3 Heh... -h is already in use...
new af02f92 Safer handling of strings which may have spaces in them.
new 9ee059d Missed one handling case
new 7517e85 Fix floating point precision error which was causing issues only when compiled with optimizations. I suspect this to be a manifestation of http://gcc.gnu.org/bugzilla/show_bug.cgi?id=323 and making one argument volatile is proposed as a potential simple solution, which appears to work here.
new 23ca8a6 Use base of 1.3.
new 0065ddf Refactor and try to rewrite dual tree traverser so it is faster, but I don't think we're completely there yet.
new 7012823 Comment out debug output for now.
new 1ab68db Comment out some debugging output.
new 84bc349 Update bounds -- the tolerance was being exceeded (just barely).
new 353a3a2 Update documentation for the use of cover trees.
new debe167 Change expansion constant to base (#240).
new f86dda7 #240: Change expansion constant to base.
new 28e2630 Must used square rooted distance to be a valid metric.
new 16888a4 Use valid alternate metric in tests.
new 323d706 Fix a warning which was actually also a bug; Score() takes the base case evaluation as a third argument (optionally), not a previous score.
new 70b7ac8 Update copyright file (should this be in a different format?).
new 24259a7 New contributor
new 35d1571 noinline wasn't suppoed to be checked in, oops.
new 6faeca3 Add one more contributor.
new de8d517 A few more modifications to the tree test
new 88937cb Propagate r13418 to trunk.
new 27e1c62 Trivially minor misspelling.
new ab611e5 #164: Any objects passed into a prefixedoutstream that contain a std::string ToString() const method will write the results of the ToString method instead of simply being passed directly into the prefixedoutstream.
new f6860f6 #164: Started documenting sfinae_utilities.hpp.
new b5526f2 Added more documentation to sfinae_utility.hpp.
new 79c452b #147 Cleaned up code. Added some documentation. Implemented ToString() for GaussianDistribution.
new 9cc2ff8 Add inplace_reshape() for reshaping matrices without doing anything to the data.
new c97c0cb Add a test for inplace_reshape().
new 4c6e7d4 Document the impending doom which will certainly be caused if you use this function carelessly.
new f510ea7 Complete sentences, yo.
new 9507271 Fix odd spacing and be unnecessarily pedantic with const-correctness.
new 4c82758 That empty initalizer is implicit anyway.
new 6d79880 Add constructor which automatically creates identity covariance, and don't check if covariance is initialized when calling Evaluate().
new eb171b7 Add tests for new MahalanobisDistance constructor.
new 645ce14 Documentation fix, and tabs to spaces.
new 6189853 Documentation fixes.
new b3e1bce Change template parameter names to be more in accordance with the rest of the codebase.
new 5f4f638 Change template parameter names to be more in accordance with the rest of the codebase.
new fb078e0 Add TakeRoot parameter for use with LMetric<n, true>.
new 400d2ab Test new HRectBound<n, true> functionality.
new 66cae0b Make a stronger note that the LMetric without taking a root is not a valid metric.
new 7a8b730 Update documentation for the HRectBound.
new 278da0a Ack, that's an annoying typo in a bad place.
new f176042 Another annoying typo.
new bf0de6a Update some documentation for clarity and coding standards.
new cad3212 Longer lines than 80 characters? Unthinkable!
new 48cf155 Debugging output, not warning output. Oops.
new 0dbde50 Fix grammar in comments (utter pedantry) and a few style issues.
new 7cd2d54 Tabs? In my application? It's more likely than you think.
new 70633a3 First cleanup of code: adhere to formatting standards (mostly).
new 9da4775 The mean was not being set to zero in the constructor where the documentation said it was being set to zero. This is now fixed.
new 9bbbe62 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@13676 9d5b8971-822b-0410-80eb-d18c1038ef23
new 278db94 Remove test, because it's just testing internal MLPACK functionality which is already tested.
new 2f16759 Fix includes.
new 0ee4969 Updated EMST code to use tree traverser abstractions.
new b3a0dc8 Utility to find MATLAB and the mex compiler.
new 03f038a Build rules for MATLAB bindings.
new 1ef239a Build rules for the EMST MATLAB mex binding.
new 8b05378 Makefile no longer necessary.
new e44f016 Finish CMake configuration changes for MATLAB bindings. This *should* break the build (and will test what FindMatlabMex does when MATLAB isn't installed). I'll have it fixed shortly thereafter. Yes, this is somewhat lazy testing...
new 236a981 This comment is no longer necessary.
new 68ad657 This comment isn't necessary.
new 0f79e11 Actually issue errors when mex is not found.
new 51b96a8 Clean up script a little bit, make it a bit friendlier.
new ddafd1f Clean up configuration so that MLPACK will compile on systems without MATLAB. Also handle CMake policies more gracefully; I can't believe CMake just doesn't ignore policies it doesn't know about. Poor design...
new abcc605 Remove all the cmake_minimum_required() bits that weren't actually necessary because there is a cmake_minimum_required() in the root CMakeLists.txt.
new 4d237c4 Further clean up script and handle things better when MATLAB does exist.
new 184d649 Use MATLABMEX_FOUND not MATLAB_MEX since CMake string comparisons are as stupid as my cat.
new 44bad79 Comments and cmake_minimum_required removed because they are not necessary.
new 1ea435a Script to modify the file which controls the MATLAB default path to add the "MLPACK toolbox", as I am calling it. Also in reference to an earlier commit message, to clarify, my cat is quite seriously weapons-grade stupid.
new 9a5ed33 Add MLPACK MATLAB bindings to the default MATLAB path.
new a28b366 Add new developer (Patrick did the MATLAB bindings).
new 9292767 Update copyright
new 2f1d361 Add DETs to the list.
new 20b5960 Change name of output and update MATLAB binding accordingly.
new cb41bd0 Tabs? In your codebase? It's more likely than you think.
new 46df87b git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@13752 9d5b8971-822b-0410-80eb-d18c1038ef23
new c6bef0d Slight reorganization.
new 272c5c8 Weird misspelling fixed
new e690690 Stochastic gradient descent will live here.
new f657d88 Clarify documentation. L_BFGS is a minimizer.
new 3ed99cb Not yet tested -- but implemented in a few minutes. Turns out SGD is pretty straightforward...
new 80f2a62 Modify namespace, and create CMakeLists.txt.
new edd1fd3 Add SGD, and add functionality to LBFGS test functions (which will probably be removed later).
new cc5858e Fix a few errors in these files.
new f6db945 Add a test function and make it work right.
new 3222af6 Add a test for SGD.
new f7307f8 Move everything into the mlpack::util namespace from mlpack::io or mlpack::utilities.
new 6829bf9 Update to use mlpack::util namespace.
new 3e39c65 Fix formatting issues and prepare for new functions.
new 892c708 Fixes to use mlpack::util not mlpack::utilities.
new fdedbec Implement decomposable error function and gradient, then make NCA use SGD and not L-BFGS. That's a lot of acronyms... for now, parameters to SGD are specifiable through the NCA interface, but I think in the long run NCA should be templatized to use an arbitrary optimizer.
new 7e617fa More parameters for NCA in accordance with the change to SGD.
new b743a28 Allow normalization of points, to prevent underflows.
new e14b1b4 added range_search and gmm; CMake updates for kmeans
new 8477b8d Remove the normalization stuff... that is not the right way to do it.
new 52c1e2b Slightly smarter normalization strategy for large datasets.
new a677184 Also support a random seed.
new ba2f8a7 Stupid, stupid, stupid (#256)
new 142b566 Oops, I missed one utilities -> util change.
new db20dd8 Explicit cast
new d671419 Better default tolerance; 1e-5 is too high and may "accidentally" converge too soon.
new 370ae47 We were neglecting a term in the gradient...
new 19f6d46 Tests for the separable evaluation and gradient of the NCA softmax error function.
new aa724f1 Add parameter to determine whether or not SGD is shuffled.
new d642272 For real, I seriously considered anger management classes while hunting this one down.
new 41a1c4a Allow passing an initial matrix.
new bf00af5 Change the way things are normalized.
new 7174b79 No need to print the final matrix.
new e3877da A better normalization strategy. And actually document how you make NCA work well.
new d6091bb Output the size when loading. How cool!
new 4f2c363 Rename in accordance with the submitted paper.
new f296da0 Update include location.
new fbd980d Rename all MaxIP files to FastMKS.
new eb6b9fd Rename MaxIP test to FastMKS test and update everything accordingly.
new a6fa1f2 Remove sparse support from trunk.
new 338e030 Oops, don't remove inplace_reshape().
new 028b03d Only do sparse tests if sparse matrix support exists.
new 3b2ad36 Add explicit constructors for sparse vectors, which are applied only when Armadillo 3.4 is used (this is a backport of new functionality).
new 8419c4c No need to typedef u64 anymore; set size_t == uword by defining ARMA_64BIT_WORD on systems where size_t is 64 bits; clean up preprocessor #ifs so that Row_ and Col_extra_bones work correctly.
new 8899373 Update CMake configuration for removed files.
new 44ce832 Fix links. #249
new ad77974 Workaround for the case of sparse matrices.
new e8f050e Bake in support for HDF5 -- if Armadillo has it.
new 668605a Test HDF5 load/save functionality.
new 7bf913d Fix a code error, fix build instructions to include package managers.
new 5edd2db Actually, we can't just define ARMA_64BIT_WORD willy-nilly because it makes weird weird WEIRD things happen. Yeah, no joke. If you want some fun sometime... try that...
new dc26f37 Check for a 64-bit system where ARMA_64BIT_WORD is not enabled, and warn the user if that is the case. MLPACK still works fine without ARMA_64BIT_WORD, but it is not a bad idea to ensure size_t == uword (by defining ARMA_64BIT_WORD on 64-bit systems).
new 6118807 Issue a warning when a user tries to compile MATLAB bindings and they have a version of gcc which is "not supported" by mex.
new f74c307 Fix to work with older GCC versions in cases where math:: is ambiguous (stop using arma::).
new 5b36099 No need to warn if the MATLAB mex compiler is using a newer glibc than the system.
new e7815e5 Use specific doubles for RHEL5 Boost compatibility.
new 3cff764 Clean up MATLAB bindings for EMST.
new 7663851 Update AllkFN binding documentation to how it should eventually be.
new e775657 Tabs to spaces
new 16d8a70 Very minor documentation change.
new ae6b9e3 Add build rules, tabs to spaces, and remove unnecessary header files.
new 702fdde Update headers, tabs to spaces.
new ec8afa7 Remove unnecessary Makefile, change to appropriate headers, and add header comment (and tabs to spaces).
new dd8a9b6 Remove unnecessary Makefile; add header comment, remove unnecessary headers, and change tabs to spaces.
new c4821ce Remove unnecessary Makefile, switch tabs to spaces, add header comment block.
new e9a21dc Error in comment
new 6931b4c lars, nca, nmf, pca, kernel_pca
new deaa6a4 Clean up warning.
new a8c8275 Fix documentation.
new 66dddb4 Manual tree constructor
new b63200b Test for the manual constructor.
new d29b8cb Update for cover tree API change.
new 02dec27 Add option to compile MATLAB bindings.
new 1c4f367 Make recursion into the bindings optional.
new 019eab2 Add a few new bindings and make the mex compiler required if compiled with -DMATLAB_BINDINGS=ON.
new 63274bc Move conditional on matlab bindings to matlab-specific subdirectory.
new cd229c5 Put MATLAB_BINDINGS conditional at a lower level.
new 41d1d65 Add CMake configuration for this binding.
new 2ff1a1b Tabs to spaces and correct includes.
new fdd318d Remove unnecessary Makefile.
new 6077706 Remove unnecessary Makefile, tabs to spaces, use correct includes.
new 94fa95f Tabs to spaces, fix includes, remove unnecessary Makefile.
new 7861522 Fix includes, tabs to spaces, remove unnecessary Makefile.
new d8eba4a Fix includes, remove unnecessary Makefile, tabs to spaces.
new ccaa950 Add mutators for ParentDistance() and FurthestDescendantDistance() and fix a comment.
new d4613d6 Add a method to get the furthest descendant distance.
new 3f506f9 Add a test for the furthest descendant distance.
new 322dcab Calculate queryNode.Stat().Bound() on the fly instead of just one time at UpdateAfterRecursion(). This slows things down, but can probably be modified to be faster (maybe it is not necessary in Rescore()?).
new c91b91c Don't use UpdateAfterRecursion().
new 7b7e094 Update comments for correctness to resolve #261.
new f5c7737 Another comment cleanup.
new e757d3b Refactor NCA to take an optimizer as a template parameter, and default to SGD.
new 5cdf6d7 Update test for new NCA constructor.
new 32daf84 Add option for optimizer type to parameters.
new 84b0818 Add function for cover tree pruning.
new 8244ddf Add Prescore() function which can be used to prune based off of parent distances.
new 960898a Use Prescore() to prune reference children without evaluating the base case.
new c573310 Try to prune when copying reference maps.
new a9063da Add new (hackish) rule for pruning while descending queries.
new a829d1e Fix NCASimpleDataset test.
new 4ed04e7 Change formatting to fix style.
new e07d39b Clean up RADICAL code and add some more output.
new bdbc847 Clean up documentation slightly for the case of slow runs.
new dba5d67 Implemented the ToString SFINAE pattern.
new ae62930 Implemented ToString methods for the math folder.
new a5d1135 Working on implementing the ToString functionality for the tree folder - getting a seg fault with the cover tree class.
new ccfbd6b git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@14031 9d5b8971-822b-0410-80eb-d18c1038ef23
new 529a22b Update to newest version (I mistakenly branched an old version).
new 07d39bc Minor formatting improvements and warning fix.
new 8434adb Further implementation cleanup; build rules fix, and move test to test directory.
new 356cbe1 Build LSH.
new 4c9ffb3 Add tests for LSH. They may be modified as time goes on.
new 523485d Implemented tree folder ToString methods. Anything involving a mrkd_statistic, binary_space_tree, and cover_tree are encountering segfaults with when they are being destructed, though.
new 62ffc80 Abbreviated the Range ToString method to return one line.
new 1492048 Resolved tests to compile with ancient versions of boost.
new d46b6ed Standardized code style for DiscreteDistribution's ToString method.
new a7e6477 Initializing MRKDStatistic's leftStat and rightStat fields and standardized cover_tree/cover_tree_impl.hpp
new c69be37 Fix line widths.
new dff186c Do not make copies of objects when printing to output.
new b501e28 Add copy constructor explicitly.
new 3b6ffed Add test for cover tree copy constructor.
new d00fab4 Minor refactoring of BinarySpaceTree and addition of copy constructor.
new bc66908 Test for copy constructor of binary space tree.
new d8631fa Fix unsafe code in HRectBound; bounds can be NULL.
new 89a039c Refactor MRKDStatistic into three files (maybe a little overkill, I realize now). Add accessors and mutators after making internal variables private.
new 0bb7fcd Update for MRKDStatistic API change.
new d7422a3 Flush status message to output before the size of the loaded matrix is determined.
new 88bd395 Some tricks to get u64/s64 support on more recent versions of Armadillo.
new 99e921f Add parent and Parent() so that trees can be traveled both down and up.
new df16baa Add test for Parent() method in BinarySpaceTree.
new 126b607 Issue a warning (not a fatal error) if Armadillo's configuration file does not exist. The presence of that condition will probably cause errors later on, though.
new cc3b74d A few changes to PrefixedOutStream and NullOutStream to deal with references better.
new 929aa73 Refactor calculation of bound. It may need a further fix.
new 71090e9 Don't use Prescore(); it doesn't work. Comment out for now (it may be removed later).
new 6c56e16 Add two new files to build.
new 2ae4aef Better calculation of bound. We could still optimize a little bit. Debug output is for #243; this solves #264.
new 866d680 LSH class updated: Hash width computation moved within the class and removed from the main file. More comments added to the LSH class. Search function made tunable allowing the user to chose the number of hash table he/she choses to search in.
new 9bf880a RANN moved into trunk from contrib/pram. Test added to the mlpack/tests/ directory along with relevant datasets
new f127e6c test data loading fixed
new d976afc Clean up string util files (no functionality change).
new 31e9068 Oops; turns out there _was_ a functionality change...
new d8de68c LSH Test modified
new 226d863 Add traits for trees.
new 95270e3 Add test for tree traits.
new 902154e minor update to remove some warning for the pow() function
new ec704dd minor update to remove some warning for the pow() function
new db43dbe It turns out we can actually make the bound a little tighter, which I accidentally showed yesterday.
new 34e3f53 Make GMM::Estimate() return double.
new 166fab9 Update test for changed GMM API.
new e0c876a Clean up LSH code and add random seed parameter.
new 5f42dca Add LSH to listed methods.
new d11a7f9 Remove using directive and extra whitespace.
new af2388c Simple cleanups of RANN code.
new 21dbe83 LSHSearch class MetricType template removed and only metric::SquareEuclideanDistance used appropriately throughout the class
new 5f82a1c cleaned up commented code in allknn allkfn. adding the incomplete hmm binding
new 24a92bc Far less hackery is required for Armadillo 3.6.2.
new 85fbc73 Old versions of Armadillo are no longer allowed, so this code for Armadillo < 2.1.91 is unnecessary.
new 9a17130 Fix warning.
new 42a584b Fully qualified name of var() for compatibility.
new 6e8c37d Exclude Armadillo 3.4 from K-Means sparse test because sparse var() was not implemented in that particular release.
new 5f7053a LSH requires resize() which is not present until 2.4.2.
new a8574fa Got the name of the macro wrong...
new 5575ce5 Armadillo 3.4 has some sparse bugs which cause this test to get stuck in an infinite loop. The bug is not present in Armadillo 3.6. Therefore, don't run this test on Armadillo 3.4. If a user ever reports it, then they should move to a newer version of Armadillo.
new 5f81803 'force_inline' added to ra_search_rules_impl.hpp
new f461dc8 Add history.
new bc4548a Abstract away the unmapping of points. Also fix a few bugs in AllkFN because that executable was really not kept up to date with the AllkNN executable.
new 921f463 Test Unmap().
new 57545e0 Forgot to test-build allkfn...
new 67a8d28 Change rowMajor parameter to transposeData parameter, which is set to true by default. This is in line with the LARS executable and fixes a bug; the LARS executable transposed the data when it should not have.
new d94226e Update for changed LARS API.
new ef519dc Comment out checks for now; setting the random seed does not guarantee the same results across architectures (and Boost versions) and therefore isn't a good test. We will need to write the test differently (probabilistically).
new c6470c8 Clean up tests; split into four (not three) tests, remove random seed setting where appropriate.
new e5b375b Move default arguments into declaration of function to make clang happy.
new 36b326f Cleanup for clang warnings.
new 8d1958e Update style; remove _s.
new dbeac62 Add Parent() function.
new 45bc45d Update test so Parent() is also tested, and also update constructor parameters.
new 2fee7e2 Reset parents correctly in the case of implicit nodes.
new 0235b1e A few more examples to reduce the probability of error.
new 359480f By default, take the root. Then it is a true metric. (L2-squared distance is not a metric).
new 0267503 By default, take the root. L2-squared distance is not a metric.
new ef5be45 Still use L2-squared distance by default; this needs to be fixed.
new bbf9b8d Don't use L2-squared distance by default.
new 5ea9423 Update tests because LMetric<2> is no longer the squared Euclidean distance.
new 1b05397 Use the bound in the ICML paper.
new baee742 Update to only use L2-squared distances (for now).
new 5bb45b3 Simplify constructors; we don't use any of the odd ones.
new fddf5c3 Simplified statistic construction.
new dcb111c Adapt to modified statistic API.
new 4c58f1d Adapt to new statistic API.
new 5df7b62 Style changes
new 1a30763 Oops, I did not realize there was an _impl file...
new a9ce869 Update statistic API.
new 92e8743 Update statistic API; for cover trees, actually call the initializer.
new 9ec3732 Don't use start anymore.
new 5119616 Clean up code a little, and allow returning of the metric associated with the hyperrectangle bound.
new af2db5a Remove empty constructor; store dataset; allow return of metric with Metric(). We don't seem to pay an extra price for this because sizeof(BinarySpaceTree) > 64 bytes but < 128 bytes (two cache lines). Also, reducing the size to one cache line doesn't seem to provide much benefit (at least on amd64).
new f4359eb Add Metric() function, and refactor a little bit. The metric may be stored locally if none is given.
new d6a38e8 Add test for Dataset() and update because the empty BinarySpaceTree constructor no longer exists.
new ff70fa9 The name of the executable changed. Update documentation accordingly.
new fbcc1fd Minor documentation change.
new 003c0f0 Add a readme. Readmes are good.
new df91d03 Fix bug in bound function for neighbor search. It is necessary to keep two separate bounds for each node.
new a54d4f7 Add a new trait.
new 5ebdee3 Allow returning of instantiated metric and centroid.
new 186d361 Add Diameter() function, which is useful for obtaining the furthest descendant distance.
new b417457 Test Diameter().
new cefdd5d Cache furthest descendant distance.
new e376028 All definitions should come before mlpack includes.
new 4ffea72 Make FastMKS work in the dual-tree setting. Hooray!
new 716de8d Clarify documentation significantly on Train(). Only allow setting of dimensionality in constructor and by hand, to prevent unexpected behavior in Train(). Check dimensionality of observation sequences in Train().
new 321e78b Better debugging output for DiscreteDistribution.
new f0a8722 Explicit namespace inclusion.
new c9b74c1 Clean up warnings.
new 2a6820c Use custom FindArmadillo script, which will find (and link against) HDF5 if necessary.
new 95dedcf Clean up output a little bit.
new c8dac1e Update name of inclusion guard to fit with the rest of them.
new 0a4bacd Uh, this is a major problem.
new f59d36e Superfluous c_str().
new 1c69da3 Make Save() actually return a useful value.
new c791965 Add Load() and Save() functionality for GMMs.
new 88945c3 Test Save() and Load() functionality for GMMs.
new 847fe89 Use GMM::Save() instead of a handwritten function.
new b48b140 Update coding standards.
new 51bd3e9 Add perturbation to ensure that the covariance matrix does not end up empty (and therefore singular).
new d65786c Add option for number of trials.
new eefffb9 Add perturbation to covariances when necessary to prevent zero-valued covariance matrices.
new 5710007 Zeroes in any element of a diagonal of a covariance matrix can cause problems.
new e0c99e3 This should be informational output which is present even when debugging symbols aren't.
new 0a959d4 Do not divide by zero. This is important when there are outliers whose conditional probabilities for all clusters may be zero. The results are unaffected.
new 20a5e54 Too many spaces.
new f58ab19 Put in checks for non-invertible covariances after training; add perturbation to avoid them where necessary. Also avoid dividing by zero.
new edbe390 This is necessary (apparently) until the FindArmadillo changes either get committed or wrap_libs fixes the problem entirely.
new e1d8773 Parameterize perturbation, tolerance, and maximum number of iterations.
new e60458c Add parameters for EM algorithm, and also add parameter which adds random Gaussian noise.
new 32cdc6a Don't use a specified perturbation but instead a simple (but rather slow) algorithm to detect positive definiteness of covariance matrices and force them to be positive definite if necessary. All of this is "singularity avoidance", to keep covariance matrices from becoming extremely small and then introducing NaNs and infs everywhere.
new f203770 Remove perturbation parameter and allow specification of whether or not matrices should be forced to be positive definite.
new 6ba64c3 I think this warning can be somewhat misleading, so it should be removed.
new badab92 Eek, I broke the build!
new 01bd52a Use the same fix from GMMs to ensure that covariance matrices are positive definite.
new 369a7db Allow parameterization of the tolerance of the Baum-Welch algorithm.
new 0ca76ec Add option for tolerance of Baum-Welch algorithm.
new 10ddd39 Slight documentation fix.
new 907c0c6 Minor formatting change.
new 606d096 Oops, this is not the accessor/mutator design methodology we are using.
new 08d4aa7 Avoid issuing two warnings for one problem.
new a659530 Add option to get the clusters back when the process completes.
new 4753071 Wrong index, pointed out by se7en7 in IRC
new af3c578 Fix bug in error checking, and make given errors a little more descriptive.
new 9ceaecf Better error checking, and, actually *use* the initial guesses if they were given to us. Logic is hard...
new 88f2d52 Test the new functionality of k-means: initial assignments and initial centroids. Also test that k-means fails on corner case datasets -- it should.
new 92f9390 Add a new initialization method, which is an implementation of Bradley and Fayyad's scheme for better k-means initialization points. It seems to work well.
new bca7ff5 Add test for Bradley-Fayyad initialization (RefinedStart).
new 00c64f8 Add accessors and mutators, and mark function const (as it should be).
new d773bcf Minor formatting change.
new 44162f5 Remove Pelleg-Moore support (as per #251). Allow the user to specify a file to save centroids to, and also allow the user to ask for Bradley-Fayyad initialization, and give options for the parameters to that initialization.
new 53f8a74 Allow Bradley-Fayyad initialization for k-means as initialization for EM algorithm.
new 0d64f52 Oh yeah -- actually allow the user-specified parameters to *do* something...
new 75029dc I don't like how Doxygen does @see; this should look a little better.
new d6836db Fix documentation for prettier @see results.
new f04ff1a Remove ambiguous option
new 2dd624d Name of argument changed
new 7498ae7 If loading fails, don't try to continue.
new 138a32d Patch for #272, pointed out by Marcus -- incorrect example.
new 2667645 Add directory for k-means tutorial.
new 9cdfb48 Clean up DET tutorial significantly. It could still use some work -- but then, so could the actual DTree API.
new 9d92c7c Remove reference to alternate volume regularization example.
new 7db7733 The last section shouldn't be a subsection.
new 1a62589 The last section shouldn't be a subsection.
new f4ef251 The last section shouldn't be a subsection.
new 5e81990 The last section shouldn't be a subsection.
new 73ec815 output_file is the option, not outputFile...
new f1a363e Stop timers if saving or loading fails.
new 7515d17 Fix parameter names (asleep at the wheel?).
new 1978844 Change DistanceMetric to MetricType to be more in line with the rest of the mlpack codebase.
new f1bb1c4 Wow, I spent a lot of time writing that. It's mostly checked for errors.
new c35e08c implementation of the \c EmptyCluster() function makes Link to k-means tutorial.
new 2cce81c Add directory for FastMKS tutorial.
new 85155ff It all has to be one big comment to be a page...
new 7d1dd1f Refactor cover tree constructors into one CreateChildren() function, which saves a lot of space. A little bit of changing needed to be done to set the scale right in the root node, and as a result the assert that checks that the furthest descendant distance is within the scale has been removed (since that has never been a problem). This is for #274, but the other constructor will not be changed until mlpack 1.1.0.
new f2c7341 Widen tolerances on tests that seem to be failing often.
new 407289e Widen tolerance even more.
new 7cc6397 Update IPMetric API and document it.
new c6a6a75 Update FastMKS API and do some documentation. Add some new constructors for flexibility.
new eb5d018 Remove base option -- just for now -- and update calls to FastMKS constructor for new API.
new 2c0ccc2 Clean up rules file a little bit; make BaseCase forced inline.
new 1b87355 Remove parameters that don't exist.
new d2e0a63 Stop counting kernel evaluations and distance evaluations. This may be brought back but before, it was done in a very ugly way...
new cdae3b9 Modify main executable to support --base option again, and give better documentation.
new 8848253 Don't use global distanceEvaluations anymore. Remove debugging output.
new e9593ff So Doxygen picks up that comment too...
new 92f5463 Clean up polynomial kernel and reverse input arguments so they are now in line with what FastMKS actually implements.
new a3c6e33 Update for PolynomialKernel constructor API change.
new db6bb1d I don't have a reliable connection, so check this in so I can work on it from my local system.
new abb21a2 Fix reference.
new 4f14879 Fix reference.
new f4a4447 Finished FastMKS tutorial.
new 7e6f360 Add new tutorial to list.
new e044120 Fix @file lines.
new d0a35e6 Fix @file lines.
new 7bee262 Add links to new tutorials.
new 680d9a6 A notice that FastCluster() doesn't work.
new 3aad97a Update documentation (#275).
new 1fca61b Patch for #278, contributed by Marcus Edel (thanks!).
new b2fe52f Avoid inverting empty matrices. Also fix a possible uninitialized memory issue.
new c4efb80 Merge in history for 1.0.5.
new 1a71ad0 Two functions which are useful for RangeSearch.
new b395faf Add RangeDistance() hooks to HRectBound.
new c265642 Incompetence is a surprisingly accurate predictor of build-breaking checkins.
new 73d86f2 Incremental checkin so I can work from another system. Begins outline of the RangeSearchRules class.
new 3e0f941 Add NumDescendants() and Descendant() functions which will make things much easier for range search and RANN.
new 6cf8e6b Test NumDescendants() and Descendant() instead of Begin() and Count() (since those will eventually be removed).
new e8d19ed Add Descendant() and NumDescendants() functions.
new d18b8d7 Test the Descendant() and NumDescendants() functions. These tests were surprisingly difficult to write but should be rather comprehensive.
new 0846320 Fix warning.
new b768353 Add HasSelfChildren trait.
new 4ea6786 Tests for HasSelfChildren trait.
new 170139c Revamp RangeSearch as per #244. Now this works with cover trees too!
new 1c91ee7 Comprehensive test for range search with cover trees.
new 07e4152 Grevious spelling mistake.
new c2bd474 A commit from a long time ago shows I committed the Contains() function without testing it, and, well, sure enough, it's a failure. This (hopefully) fixes it.
new 130336b Finally add test for Range::Contains(Range).
new 9a44790 I'm not sure what the older test was, but it was undocumented, so I removed it, and then added my own. I also updated the authors file for accuracy, because none of Ajinkya's tests are here anymore.
new 1f341ba Fix #280 and revamp KernelPCA implementation. It should now be much faster.
new 2efee0f Add new contributor.
new 9e1f64e Update copyright.
new c54ed5d Minor formatting fixes and better documentation.
new 33c0dec Stupid const error. Did I not test it?
new 76a2ace Move IPMetric to general metrics directory since it will be useful for other things.
new a6c2c4b Use Log::Assert.
new a87b51c Move IPMetric to mlpack::metric not mlpack::fastmks.
new 13f77fd Document what the namespace contains.
new ae33f45 Fix for #289 (thanks Marcus)
new c7ecb98 Remove unused file extensions from Doxyfile.
new 6f6ee58 test commit
new 064fac4 Removing "test.txt"
new 73b0d4d Do not compile DET code on Windows because it causes the compiler to segfault. Sigh...
new 48d1b96 Yeah, well, it's not like anything was going to fix the Windows build anyway. Now it crashes in some other way.
new 6228e06 Sometimes the map is empty, and dereferencing an empty iterator results in unhappy behavior.
new fc4355c Make the NMF decomposition more accurate in a bid to fix the failing test.
new 5f95c0e Fix ambiguous math reference error, to pass the test on OS X with clang.
new 3f8e159 Fix unused variable warnings.
new 8c53bca Widen tolerances on GMM tests.
new 1000d78 Add target_link_libraries for librt to support clock_gettime() for Linux.
new f3cc139 Update the timer class, to get stable time values with more accuracy for the common operating systems.
new fc5be7a Extra cleanup of Windows' ridiculous #define min and #define max. Seriously guys?
new 906d645 Remove unnecessary include.
new 00ab4fb Remove unnecessary include.
new 48f4a75 How did that happen?
new f2fe0e2 Reorder includes.
new a200007 Windows does not like the #define private public hack, so do not use it there. Also remove it from a few tests where it wasn't actually needed.
new 26e4ff5 logNegError is private, so don't test that on Windows.
new 9654a96 Copy HISTORY from branches for mlpack 1.0.6 update.
new ef86680 Remove unnecessary include.
new 9bcd63c Fix #291, which is a corner case in situations where the dataset consists entirely of the same point.
new 44fef7b Some formatting fixes, and create a new test that tests the scaling to unit variance option.
new 4565b03 Fix the scaling parameter; what was being done before did not make sense.
new 202854b Fix ambiguous math reference error, to pass the test on OS X with clang.
new 0d02e51 Oops, I forgot to remove the random seed from the test...
new 241ebaf Fix ambiguous match: option -s matches --seed and --single_mode.
new 240e103 Fix index out of bounds error when using det with a labels file.
new 2d7f931 Make CLI::GetParam<bool> return the same value as CLI::HasParam().
new b98cb25 Test that HasParam() returns the same as CLI::GetParam<bool>().
new 70c38b0 Clarify comments and fix bug that occurs when no query set is specified.
new 0c64e49 Fix dangling spaces and use Log::Assert() not assert().
new dcccfac Test single-dataset mode. In writing this test, bug #293 was created.
new 367e237 Fix ambiguous match: option -s matches --single and --scale.
new 70402e6 Testing diff in commit emails.
new ece403d Test diffs in commit emails again.
new a4fa555 Fix ambiguous matches for -N and -L and set default optimzer sgd.
new 7a81b63 Move constructor to top of file, and check the size of the s and y cubes before optimizing (not just at construction time) because they can be changed (see nca_main.cpp).
new 25f8db7 Rename a few tests so they compile correctly.
new e8c3a36 So, actually compiling the L-BFGS test might be a good thing...
new 2c2c790 Fix ambiguous call to pow, to pass the test with clang.
new 7f6e7a7 Comment out --fast_kmeans option so it does not work at all, for now.
new ec4fc9c I have been looking for this memory leak for months. Apparently I forgot that this file exists, and once I found it, the leak was obvious.
new ba5d819 Does this fix the cover tree two datasets test? It seems to on this box.
new 5e034d8 Don't recurse into the map pruning section if the map is already empty.
new 6ca9680 Add utility functions to normalize labels.
new 39f0d43 Modify core.hpp to include label normalization methods and update build configuration.
new 5f9d843 Test label normalization.
new 8e59ec3 Remove debugging output.
new 6ff1c80 This may fix the segfaulting in the cover tree tests.
new 5176dc0 Change label type to arma::Col<size_t> not arma::uvec.
new 2ddb6e6 Change label type to arma::Col<size_t> not arma::uvec (this is more in line with the rest of mlpack's labels).
new b03d259 Update NaiveBayesClassifier API to take labels as a separate vector. Also, normalize the labels in the main executable.
new f3dd29b Uncomment checks that weren't failing in the NBC test and update for newer NaiveBayesClassifier API. Also, normalize labels before running NBC.
new 7fc77a8 Use arma::Col<size_t> instead of arma::uvec for labels.
new f5272f3 Normalize the labels before performing computation, and use arma::Col<size_t> instead of arma::uvec.
new 1743ae0 Update test to use arma::Col<size_t> for labels instead of arma::uvec.
new d8183c6 Ticket #293 initial fix -- review waiting.
new 0d76b04 Chebyshev distance (L-infinity distance) added to lmetric.hpp
new ee166c7 LBFGS test compilation failure fixed
new 5fa2c1d metric_test.cpp -- arma::sum changed to arma::accu; LINFMetricTest corrected.
new 8c61865 This should fix the failing build on old Armadillo versions.
new e493602 Oh holy crap, I am the worst, and I have broken everything.
new 3a7d043 I always say "build before you check in!" and then the simple fix I try to commit breaks everything because I didn't have time to build it or thought "oh man, this will definitely work". I bet if I searched for commit messages like this I'd find dozens...
new ab43080 Older versions of Armadillo require as_scalar() because gcc isn't able to figure out that the result will be a double.
new 4652ce0 Fix ambiguous match: option -d matches --degree and --new_dimensionality.
new 0bcb6c5 Simplify centering code. Maybe it's faster?
new 06c4a53 Clean up PCA test.
new 23c8ae3 Refactor PCA to use SVD because it is faster.
new ac6c48f Add opportunity to set initial centroids.
new 616415b Add command line option to set initial centroids.
new 6fd0998 Minor formatting fixes, make things const 'cause that's what's cool to do, issue a warning if -r and -I are both passed (because in that case -I is ignored).
new d5e19dc Spelling fix.
new b144514 Remove c_str() where possible.
new 761efb8 Remove c_str() where possible.
new 974fe81 Widen tolerance on HMM test.
new d6ce1bd Fix misleading comment.
new 0c09b07 Marcus, finder of errors, has found yet another one.
new f95f208 Adjust tolerances.
new 9c80ea4 Adjust tolerances.
new 8836acd Minor version update. This is not a very maintainable solution...
new 4db1765 added download link to build page on mlpack.org
new 94b8494 CF framework and ALS added
new fca3bb8 For now, don't build the CF package. It uses sp_mat and that isn't included in all Armadillo versions. The solution will either be to force Armadillo >= 3.4 or #ifdef around the sparse support.
new 9e37bd6 Clean up trailing whitespace, reformat a few lines.
new 66b3bf2 Code cleanup, and refactor DTBRules so it does not depend on UpdateAfterRecursion(). Now, EMST actually provides speedup when run in dual-tree mode (hooray!).
new 7b21f44 Fix typo that breaks kmeans.
new 5e0a1b9 Add a parenthesis to fix some grammer. A big deal!
new 2a9f8d3 Clean up RANN code, document tau parameter more thoroughly, and refactor RASearchRules so the file has fewer lines and less to maintain. Clarify output when --naive is passed. Remove redundant tau validity check.
new be390f7 Add --cover_tree option to range_search.
new eb32be0 Ignore the cover_tree option if the naive option is present.
new 039d75b When removing implicit nodes, ParentDistance() was not being correctly preserved.
new c3b230b Add some changes from Sumedh (#298). This improves the Predict() function and adds an error check in the executable (I moved the error check from the Predict() function).
new b849910 Add new contributor.
new ab514dd Fix compatibility with older Armadillo versions and fix warning in linear_regression_main.cpp.
new 513338e Improve centering of the kernel matrix.
new 227f5da Handle corner case where ||a|| or ||b|| is equal to 0 (and prevent division by zero).
new 096c4f0 Fix a comment.
new 7b6e4c4 Fix output, because it does not output the squared distance any more.
new 072f1e6 Update documentation for new output; squared length is not given.
new 020e8f1 Use the correct option to output the desity estimates.
new 6a18ea9 Fix ambiguous match: option -p matches --in_place and --percentage.
new 9c2b875 Add some new functions from Sumedh.
new 5939021 Add test functions for new stuff from Sumedh.
new 8bfa973 Add -V option to specify how much variance to retain.
new 8e3b054 Add lastKernel and lastKernelNode for pre-emptive child pruning.
new fa3883a Add KernelTraits, a useful template class that can tell you about things, like whether or not your kernel is normalized. This is useful for FastMKS, and will probably be useful for other things later too.
new 329394a Just include all the kernels, because I mean, why not?
new 78e0a9a Add a test for the KernelTraits class, and a long rant which explains why this checkin didn't happen two hours ago.
new 28ccb9d No need to transpose matrices.
new c20782d This was the version of code used for the FastMKS benchmarks in the recently submitted paper, "Dual-tree Fast Exact Max-Kernel Search".
new 2eab3fb Print base cases and scores as output even when not debugging.
new b8f56c3 Often number of prunes can be useful information, to make sure that anything at all is being pruned.
new 59c9c76 Uh, this should fix the broken build. I will clean this up in the forthcoming days. The traversers are re-implemented to make the overloads of Score() and Rescore() that take base case values unnecessary, and actually the dual-tree traverser now satisfies the definition given in the ICML paper correctly.
new b5be225 Add functions to get number of base cases, number of node combinations visited.
new 161b794 Overhaul NeighborSearch so that it only needs one overload of Score() and does not pass around the base case. This fixes the failing tests with cover trees. Also, QueryStat is split into neighbor_search_stat.hpp and renamed NeighborSearchStat.
new 08c61c2 Update test for new NeighborSearchStat class.
new dd4994f Two things: actually do some pruning, and, get parent base case evaluations correctly. I cannot believe how long it took me to track that down. Ugh...
new e011b28 Add RangeSearchStat class.
new 035086b Move to right namespace.
new f784d63 Use RangeSearchStat (although nothing is being done with it yet).
new c98a7fd Use RangeSearchStat in cover tree objects.
new 67bd751 So, to test this file, you have to type 'make range_search'...
new f263f75 Adding Cosine Trees
new 2f1847a Refactor test to clean tree statistics before running new range searches.
new 808965b Refactor RangeSearch to work properly for both cover trees and kd-trees with the new cover tree traversal.
new 3020a10 Fix comment.
new a99aac4 Use RangeSearchStat with cover trees.
new 13451e3 Added tests for CosineTree and CosineTreeBuilder
new 1734010 Added tests for CF Module
new df07dbb r15817 overwrote the current tree tests with ones from mlpack 1.0.4. This reverts that change, and reformats the cosine tree testing code.
new 2e4fd4e For now, don't build the CF test. This is because CF depends on sp_mat, which is not available in Armadillo <= 3.4.
new adffb37 Remove EuclideanDistance() and CosineSimilarity() functions; instead, use CosineDistance from kernels/. Also remove trailing spaces (thanks, vim!).
new 23fc679 Converting row-major to column major
new a656b6b Removed an older version of tree test and added a new one
new 7eb3834 Fix compilation errors and merge actual changes from r15826.
new 85ba009 The parent-child pruning bounds were too tight. They did not consider one of the bound terms and the cross-term, and thus failed, especially when considering a leaf node where the furthest descendant distance was 0. Also update documentation a bit so that big chunk of code makes a bit more sense.
new 9b1ec56 Update tutorial.
new 90119bf A few fixes.
new 2ceb65c A few more fixes.
new fbc568d Hey look, I removed three lines of code while not changing the functionality of anything. Convenient!
new 0d24a67 Use the option to set the leaf class table file and unify method call.
new 0c18387 Integrate ComputeCost() as ComputeError() for #298, written by Sumedh.
new 29431ca Add test for ComputeError() for #298.
new 3a65737 Use Sumedh's dataset for the perfect fit test.
new 4cf182a Interesting that this compiles even though the sum of a dot product is a meaningless expression.
new 8aa8551 Fix warning, add const.
new 18db53e Make Predict() function const-correct.
new 48e5b43 Widen tolerances.
new af90b56 Very minor syntax fixes.
new 2d378ee Add lambda parameter, so this now supports ridge regression.
new b321e1c Use an empty dataset, like Sumedh suggested, to test ridge regression.
new 8565ebc Avoid an additional matrix copy (insert_rows() and shed_rows() are a copy each time). Also make the input matrix const.
new d28461d Fix other constructors so they play okay with lambda.
new c29a4af Update the timer example call and output.
new 9d93619 Minor typo.
new 794cd3b Fix in implementation to avoid copies.
new 5b9eafa Add --lambda option to linear_regression.
new 0dfe581 Update documentation.
new 9a58bd0 Add section on ridge regression.
new d5f10c7 Add a note about running a timer multiple times.
new a48c8bc Update NMF to work with sparse matrices also.
new 19a3b9e More fixes for correct sparse NMF.
new b12fd2e Test sparse NMF.
new 0dbc295 Armadillo 3.6.0 will be required for 1.0.7 because that is the first version where sparse matrix support was actually available.
new aa74129 Update documentation to reflect Armadillo 3.6.0 requirement.
new a5c1b64 Update documentation to reflect 3.6.0 requirement.
new 484328d Because we now require Armadillo 3.6.0, we can use each_row() and each_col() (thanks Marcus for the reminder).
new 84f445e Fix comment headers for doxygen, remove trailing spaces, and use NMF instead of ALS because it now supports sparse matrices.
new bad46f9 Instead of using ALS, use NMF with a sparse matrix.
new 8c7a513 Remove als directory.
new e54fd5a Reformat GetRecommendations() function and rearrange cf.cpp to reflect ordering in cf.hpp a little better.
new bc5a748 Why not call CleanData() in the constructor instead of repeatedly in other parts of the implementation?
new 34f58a0 Clean up CleanData() function. Don't use temporary matrices; there's no point to doing that.
new f09c3a2 Fix ambiguous call to RandomSeed, to pass the test with clang.
new e5ab291 Minor comment fix.
new e8d9a0d Use sparse matrix batch constructor.
new 903c32f Clean up Decompose().
new 3276236 Decompose() is only called once, so just inline it because it's simple.
new 46f5cf4 Remove GenerateRating() function because it's only one line and only used in one place.
new c31639a Remove CalculateApproximateRatings() since it is only ever called once.
new 76c7b30 Clean up Query(). It makes me feel a little gross to be so pedantic as to change 'neighbourhood' to 'neighborhood', and maybe I'm taking "code consistency" a bit too seriously, but it's too late and I've already done it.
new 6a96f77 Inline CreateQuery() function since it is only used once, and use better Armadillo functions to make the generation cleaner-looking (probably no speedup).
new 35cf431 Remove GetNeighbourhood() because it is only called once. The check that numUsersForSimilarity < numUsers is redundant because AllkNN will make that check.
new 4713300 Remove CalculateAverages() since it is only called once, and inline it into Query(). The code has been cleaned up a bit and should run faster now because it does not use a temporary column vector.
new e4dff4c Refactor CalculateTopRecommendations(), including a complete overhaul of how recommendations are actually calculated. std::pair<> and std::map<> are often quite slow, especially in that implementation. This is faster.
new 1b6e5ad Issue a warning if the correct number of recommendations could not be provided.
new 7cecc02 Remove CalculateTopRecommendations() function, because it was only called once. The 'recos' variable, as best I could tell, was not used for anything.
new 047a5aa The very large and dense 'mask' matrix is now unnecessary.
new 986ce4c No! The spaces were WRONG!
new b6bcf55 Remove Query() function as it is only used once; inline it into GetRecommendations().
new 5b67190 Remove method that allows modifying the data matrix, because references can't be re-pointed, and also because the sparse data matrix is only built in the constructor.
new a082399 Clean up CFTest, and fix some comments that were wrong.
new 877171f Add convenient function to get sparse matrix.
new 5cf4672 Add a ridge regression test case (huh, I wonder how I forgot to check this in earlier).
new 6aa3c93 Add a test to make sure the output of the CF object is reasonable. I think this is a good test...
new 0e0c17d Clean up main executable; a bit of const-correctness pedantry, and don't use c_str() if it isn't needed.
new 50bb8fd Fix comment (thanks Marcus).
new 7f118a4 Build the CF test.
new e70776e Build CF module.
new 388e39f When --query_file is specified, only give recommendations for those users.
new a78f293 Remove options for output files that weren't being used, and comment out options that aren't implemented yet.
new 95c138c Better comments for unimplemented functionality.
new 6905e3e Actually use number of recommendations and size of neighborhood in calculations.
new eccc1d9 More comprehensive documentation on options and input/output formats.
new 23341ef Refactor EMFit<> so that the covariance constraint is a template parameter, and update other components accordingly.
new 81a3e19 Make function static.
new cfb9554 Add tests for NoConstraint and PositiveDefiniteCovariance.
new b3d1886 Rename PositiveDefiniteCovariance to PositiveDefiniteConstraint, because I think that's a better name and nobody's going to stop me (...I think?).
new 1e40889 Update name of constraint class.
new a43e792 Oh hey, I have to check this in too.
new 578f0dc Add DiagonalConstraint.
new 0f0fed7 Add test for DiagonalConstraint.
new 055af4c Add EigenvalueRatioConstraint class for EMFit.
new a1dbd9c Add test for EigenvalueRatioConstraint.
new d9c2c6b What, how did I miss this? I definitely built the test and checked it...
new 73d6adb Remove fake functions that should only be specific to the kd-tree.
new c65afd1 This is premature; wait for this until later.
new 325d80d If the root node ends up having an implicit child (not an illicit child!), remove the implicit node.
new f528e01 Remove unnecessary functions that were originally for kd-tree compatibility.
new 72e4f86 Handle cases where the reference map is empty a little better.
new 198a8f3 Overhaul RASearchRules so that they will work correctly with cover trees.
new 3069fdb Test RANN with cover trees.
new 678ee06 Add a note that this does not work.
new c038343 Do not compile MVU.
new 6da2e91 Fix some comments. No functionality change, and also this was probably pointless.
new 0928a53 Yeah, it makes a difference to inline the BaseCase function (#281).
new d2c7108 Force inlining (#281).
new eefddd0 Backport batch insertion constructor to Armadillo < 3.810.0.
new f02d8cc Remove backporting for versions earlier than 3.6.0 because that's just not necessary anymore (3.6.0 is the minimum Armadillo version).
new 16939b5 Oh, right, update the least maintainable part of the entire build system that I haven't figured out a solution to yet. Argh...
new d767b3d Update copyright file to include new contributors.
new be2f994 Update citation.
new 765487c Update changelogs.
new 172be16 Minor updates for normalized kernels.
new 8b6c954 One more thing...
new 32a27dd Update to latest version number.
new 4b4f09d Widen tolerances slightly.
new 00a5f42 Move DTBStat into a separate file.
new 567b0da Check the matrix dimension before dropping unnecessary rows.
new d3ff55c Add Epanechnikov kernel for kernel PCA.
new d9c3a8c Allow GMMs to be trained using the existing model as a starting point. This could probably all use some serious refactoring now.
new 984a8d9 Add tests for estimation with the existing model as the initial point.
new 8a9c715 Add logistic regression directory.
new 335045b Initial commit of logistic regression by Sumedh Ghaisas (#305).
new 78b9fa0 Add some comments and fix a little bit of formatting.
new 42a03ca Rename LogisticRegressionFunction to LogisticFunction.
new 871799f LogisticFunction is incorrect; it's actually the objective function for logistic regression, which is not the logistic function. Also remove unnecessary overloads in LogisticRegressionFunction.
new 698b2bc Refactor Evaluate() function so that it works alright, and eliminate unnecessary parameters to the function. Move LogisticRegressionFunction implementation into a .cpp file because it is not templatized (for now).
new 06bb2d3 Add test for LogisticRegressionFunction::Evaluate().
new ff1966e Oh yeah, this is necessary...
new 15c3d75 Add a randomized test for the logistic regression likelihood function that is not just comprised of simple examples.
new 40014dd Add test for regularization of objective function.
new 12db4ed Add test for Gradient().
new 7893e52 Add correct implementation of Gradient() function.
new 7f8e28e Make Gradient() const and add comments.
new 1961f9a Add some comments.
new b7cbf6a Remove getSigmoid() and add class documentation.
new 65e175e Break Predict() instead of the whole build.
new 168fd7b Revamp Evaluate(parameters, i) for SGD and related optimizers. NumFunctions() also needed to be fixed.
new c13fbc9 Add test for Evaluate(parameters, i) for SGD and related optimizers.
new d9a32b0 Test regularization for separable Evaluate().
new 6fd7543 Update Gradient() for the separable case.
new 4829cc7 Test Gradient() in the separable case.
new ca88a14 No need to multiply by responses[i] or (1 - responses[i]) because those quantities will always be 1.
new d348140 Add test for regularization in Gradient().
new e6a2271 Add a test for the separable Gradient() function.
new 0b823db And this is why I hate maintaining documentation for things that I don't control; it goes out of date really quickly and people get confused...
new 8795626 Remove memory leak (#310). Thanks to bianjiang for pointing this out.
new b98df60 Don't print "Iteration X of 0."; just "Iteration X".
new f67281f Refactor functions so that they do not expect the extra column of ones for the intercept term.
new f3d6ef3 Refactor tests so that they do not give the extra column of ones for the intercept term.
new a070c32 Make predictors and responses const.
new f819772 Make predictors and responses const.
new 022dc4b Don't hold lambda in LogisticRegression because it isn't necessary. Also make predictors and responses const because we don't need to modify them.
new 950b81f Refactor LogisticRegression class. LearnModel() can be private, and we can const-ize a lot of the parameters and functions. Also restore functionality to Predict(), ComputeError(), and ComputeAccuracy(). This does not ever make a copy of the predictors matrix and doesn't add a column for the intercept, which is faster.
new 8c3e1a0 Add tests for actual LogisticRegression object.
new bd7e239 Goodbye, slightly misleading documentation.
new 561e64d Add function to get the function that is being optimized.
new 4aa1314 Add function to get predictors and responses.
new 72374fa Add a constructor that allows passing an instantiated optimizer.
new 944e8ec Test constructor that takes an instantiated optimizer.
new 1cba5ab Update history file incrementally, because doing this all at once at release time is difficult.
new 421d5dd Add FurthestPointDistance() function.
new 11389df Test BinarySpaceTree::FurthestPointDistance().
new c79f19b Implement FurthestPointDistance() for cover trees.
new f433dd8 Regularization gets subtracted, not added. Thanks to Sumedh for pointing this out.
new 57ae1f9 Update test for positive regularization... oops...
new 373a197 Clarify comment.
new 62ec825 Remove predictors and responses because they don't need to be stored by the LogisticRegression class, and then add a constructor so you can pass a given set of parameters.
new 364e896 Don't hold the optimizer and error function since they are only needed at construction time.
new 7c94d4f Add main executable for logistic regression.
new 5123690 I don't remember doing this, and I have no idea when it was, but hey, using namespace std in cpp files is a cool idea.
new 2a98956 Add information so a version can be obtained easily during runtime and also at compile time.
new 07cd6c6 Some documentation on versioning information.
new 535786e Oops, a typo.
new bad554e Link to tutorial on versioning information in core.hpp.
new 648a84b Fix typo: furthest <-> nearest.
new dcc1b8b Update so that it gets compiled into libmlpack.so.
new e643278 CMake script to create svnversion.hpp, if needed.
new af0a03c CMake scripts to find and update svnversion.hpp, if necessary.
new b996908 Fix dependency on stringstream.
new 533293d Safe check for if we are in an svn repo.
new b07bbf7 Remove redundant check for working copy.
new ebe3b1f Add svnversion.hpp to list of sources to track, if necessary.
new 0569ee0 Fix error so that core.hpp is properly included.
new 815154c Ignore svnversion.hpp.
new aff75c6 Update copyright for the new year.
new 0055ea1 Add Mudit's name to the list of contributors.
new c75fdcf Increase precision of saves.
new f857b8c When using the constructors that don't specify a model, make sure we create an actual valid GMM and not just random memory.
new 1300e4a svn:ignore --> 'build*' in case you have multiple build directories.
new 058ec36 This wasn't actually calculating the variance correctly.
new 1dbdf57 Handle cases where scales are 0 correctly.
new 791d88e Pull in updated history from version 1.0.8.
new e2f936e Ticket #314 resulted in a bunch of useful changes.
new da5727c Fix from Michael Fox for #242.
new 43f1db6 Add tests when the data matrix has linearly dependent features.
new feb4f90 Add new contributor.
new 0f3d893 Fix spacing.
new 76d029d Added ToString() to core/dists, core/kernels, and core/metrics Added a test for the ToString() methods
new 1995b0c Updated Test for working optimizers
new e1c7c55 Added Tostring to SGD.
new e98a940 Addes Tostring here early, mostly because SGD required it.
new 43ff56d Changed so that "dimension" is not declared, to avoid compiler warning.
new 3a75faf Changed so that "numPrunes" is not declared, to avoid compiler warning.
new 5b18bba Added L_BFGS
new 944e800 Added ToString to LBFGS
new eccf54d Fixed AugLagFunction Test; working on AugLag proper.
new ad85909 Added ToString method
new 718f9c1 Added ToString for AugLag Itself.
new a00daa0 Fixed AugLag Test
new b1f958a Fixed my own error in ToString of AugLagrangian
new 654fed8 Fixed Test Error
new 0858f70 Add CMake configuration for LRSDP, and split it properly into cpp/hpp/_impl.hpp files.
new 7e14740 The Problem was the Test; Added ToString to AugLag's Test.
new 963158b Added BallBound ToString test.
new 8f210eb Don't make an extra line when indenting things.
new 5c9af5e convert << util::Indent(probabilities); Minor modifications to ToString()s in distributions.
new 4c4e884 Add new contributor.
new e6f8646 Update ToString() for all kernels except the PSpectrumStringKernel.
new 27853e3 Update so filename is like other filenames.
new c968db4 fixed ticket #316 (hmm::Predict() in hmm_impl.hpp)
new eee8865 Remove license, that doesn't get applied until a release (until then it's implicit, yo).
new df138ed Inline non-templated constructor.
new 7e06575 Update documentation (thanks Yubao).
new f321f84 Fix documentation (#321); from Siddharth.
new 45d5020 Fixed Positive Def Constaint under gmm; previous error was "/gmm/positive_definite_constraint.hpp:28: error: expected primary-expression before ‘(’ token" at lines 27 and 33.
new 3a227e8 Added ToString to GMM
new 52b6b22 Some ToString added, some updated (typos and indent). Test updated.
new 1eef878 Fixed my subversion error, hopefully.
new 29ddcea Fixed a Mistake in HMM's ToString
new 63a6f66 Added and Modified more ToString methods.
new ad83624 Fix collaborative filtering test (#323).
new c3b1d4a Add a boolean specifying whether or not the dataset is rearranged when the tree is built.
new b839382 Cache the distance from the center of the node to the center of the parent node.
new f9aef70 Test BinarySpaceTree<...>::ParentDistance().
new da95379 Also test ParentDistance() in the case where the tree is built and mappings are used.
new be55426 TraversalInfo, a new idea for doing things. This will need to be further documented...
new 77b2e69 TraversalInfo class used by nearest-neighbor-like problems. Maybe it could be the same as the one in src/mlpack/core/tree/.
new a843a89 Change build to include ns_traversal_info.hpp.
new 9e25c61 Add traversal_info.hpp to build system although I don't think it's used anywhere.
new 5272fb2 Make rules classes for various dual-tree algorithms support (but not use) the idea of TraversalInfo classes.
new 8a799d7 Minor changes to SortDistance().
new d889f08 Overhaul NeighborSearchRules to work correctly with TraversalInfo objects. This is related to #243.
new 3d04d0e Modify SortDistance() usage because SortDistance() API has changed slightly.
new 6b5733c Modify BinarySpaceTree::DualTreeTraverser to properly handle TraversalInfo objects.
new 997efc7 Modify CoverTree::DualTreeTraverser to properly handle TraversalInfo objects.
new d6b0eb0 Adapt SortPolicyTest for new API.
new f64d795 Remove unused score variable; this fixes a lot of warnings.
new 4926fce Modified StringUtil to take a number on Indent, implemented LSHSearch ToString, updated Test.
new fdcbea8 Added more ToString, updated test, took advantage of new tab function.
new 986626f Patch from Siddharth for #324.
new 3f9e7b5 Finished toString of the methods folder ( sparse coding and rann ); updated test.
new be82046 fix forHMMTest/SimpleDiscreteHMMTestViterbi
new f28995d Remove license text.
new 96c2ae2 Slight code formatting modifications.
new c5e9f79 Don't initialize empty distributions.
new c47c355 Remove unnecessary newlines from after mean and covariance output.
new 7731a5c Actually print the kernel.
new 19b21fe Minor fixes to ToString().
new 1670352 Minor fixes to ToString().
new 6226f37 Minor changes to ToString().
new e17583b Minor changes to ToString().
new 7a344fb Minor fixes to AugLagrangian ToString() methods.
new 4ee6d99 Move things around so LRSDP can actually be included in stuff.
new 3a81582 Some updates to what has been going on (lots of things!).
new 502c36b Ah, right, one more thing.
new 48e6ea5 Minor changes to ToString() output.
new 82940cf Minor changes to tests; in some cases matrices/vectors are now initialized so they provide more output.
new c575385 Remove PeriodicHRectBound (#30).
new 2c7cf54 Remove this file too...
new 4e3d03c Remove PeriodicHRectBound from tree tests.
new baff477 Denote that we removed PeriodicHRectBound.
new 31f163d PeriodicHRectBound is removed (#30).
new e4a1c23 Updated Test and pspectrum_string_kernel
new 156a775 Updated Test and some info for ToString.
new a205f41 git-svn-id: http://svn.cc.gatech.edu/fastlab/mlpack/trunk@16273 9d5b8971-822b-0410-80eb-d18c1038ef23
new 2b02777 Revert commit by unknown contributor.
new b7ac346 Test commit
new 8a836b3 Revert test commit.
new 3403a83 Update citation format.
new 9bb6000 Fix from Siddharth, fixes #324.
new d488300 Add new contributor.
new 764a1c1 Fix formatting and fix contributor information.
new 3cca93a Didn't mean to check in arma_traits.hpp...
new a33ea2b Oops, forgot to check in this change from Siddharth.
new 815d470 Example tree. Not really to be used, but for documentation.
new 528809f Refactoring from Saheb: don't do naive search with trees.
new b86d4fd Add new contributor.
new 49aa1eb Clean up RangeSearch implementation so it works with TraversalInfo struct.
new 7bb869a Use TraversalInfo for FastMKS. Right now, the parent-child prune is not implementd, so this is significantly slower than the released version (but it at least works, unlike the last svn revision).
new cb763a1 When the parent-child rule can't be applied, the correct bound wasn't being used before. This fixes that.
new 677f1d8 Do child-parent and parent-parent prunes correctly for FastMKS. This actually accelerates the dual-tree implementation, which is cool. By how much? Not really sure, maybe 10%-20%?
new a108498 Update the line about dual-tree algorithms being faster. 'cause they are.
new 77250b9 The Prescore() and PrescoreQ() functions aren't used anymore because I figured out how to do child-parent and parent-parent prunes without them using the TraversalInfo object.
new c4f91a4 Patch from Saheb: do actual naive search for RangeSearch and NeighborSearch instead of using the hack for BinarySpaceTree with leafSize = n_cols.
new 4ed8e73 Add some Armadillo traits for template metaprogramming. This abstraction may change as time goes on.
new df79aaa By default, include arma_traits.hpp.
new 600d33f Refactor MinDistance(), MaxDistance(), and RangeDistance() to accept any Armadillo vector type.
new d3f8b71 Refactor MinDistance(), MaxDistance(), and RangeDistance() to accept arbitrary types of Armadillo vectors using IsVector<>.
new 3979dab Having an implementation for the example tree is kind of a dumb idea. Also make sure CMake knows about example_tree.hpp so it keeps it up to date.
new f47d673 Clean up some namespacing, and add a test for sparse AllkNN using kd-trees. Also add one for cover trees, although it does not work yet.
new 263b5e0 Templatize arguments for BestPointToNodeDistance().
new 5d6bb9b Refactor NeighborSearch so it works with arbitrary TreeType::Mat types. That abstraction should probably be cleaned up a bit.
new 0653608 No more implementation.
new ca45205 Stop timers correctly when Log::Fatal is called (or when Load() or Save() return early).
new 7c316eb Rank is not yet a parameter, but this heuristic from Siddharth improves upon the previous "rank = 2".
new deecafd Fix mixup of rows and columns.
new bf508a2 Widen tolerances ... again.
new 1db619f Patch from Siddharth: make the rank parameterizable, and merge the constructors into one.
new 4e51ea9 Update for changed CF API (patch from Siddharth).
new fcf5dee Some changes to LaTeX options in the Doxyfile.
new c1b8faf Explicitly use namespace mlpack::neighbor.
new 3ce28f0 Remove using directive that has been there for six years. Sigh...
new 669f91a Remove numPrunes -- it isn't used.
new d4c2373 The maxLeafSize parameter is unused (other than an assert) so remove it entirely.
new 4a32b1c Initialize uninitialized variable (#336).
new ea76866 Remove numberOfPrunes variable, and initialize baseCase (#336).
new 69cdcd5 Patch from Siddharth: templatize CF to accept arbitrary types of factorizers.
new d0ed633 Fix regression in r16308, which slowed down cover tree traversals by potentially up to 50% (at least, that's the slowdown I saw...).
new c54bed8 Actually use the right parameter name.
new d67356a Ok, handle NaNs correctly, and also check this in in trunk, not in the tags...
new e96b008 Remove incorrect comments.
new 46796ca Better rule, which forces the score to 0 for AllkNN. Not sure if it'll work for AllkFN. Also, I think there may be a bug in the traversal info handling for the BinarySpaceTree DualTreeTraverser.
new d7ad8de What was I thinking? This is a _far_ better solution than r16359.
new d528eda Modified patch from Saheb for #301; this unifies the constructors for NeighborSearch so they work with all tree types. The modifications I've made make it so that the referenceCopy and queryCopy matrices aren't full copies of the referenceSet and querySet matrices when the tree doesn't modify them (in the case where they aren't modified, it's not necessary to copy them, that's just a waste of memory).
new 9933b43 Fix to go with patch from Saheb... forgot to check this in earlier.
new 46da253 Make the ExhaustiveSyntheticTest use kd-trees with a leafSize of 1. This changeset brought to you by the magic of regular expressions, which mostly worked...
new 3010bd9 Oops, use single-tree for the second iteration.
new 7626468 Through the magic of regular expressions, I've refactored this test to use kd-trees with one-point leaves.
new 0fa3501 Fixed arma_traits by removing duplicate template instantiation, so that visual studio would be able to use it too.
new 5c612b9 Update documentation to reflect changes in RangeSearch.
new f84c023 Incorporate patch from yashdv to move splitting procedure to a different class.
new 22c7af5 Add new contributor.
new 5363972 Make parameter const because const-correctness is a good thing.
new ce1bcd2 Don't use the SquaredEuclideanDistance.
new b100fd6 Don't use the squared Euclidean distance.
new e480941 Fix minor error that I didn't correct last time.
new 4a11ed1 Refactor LRSDP into the main class and a separate function (LRSDPFunction). This is a patch from Abhishek Laddha for #318.
new f6084b6 Better ToString() output.
new 8c022cf Annotate changes to LRSDP.
new 4177afa Add new contributor.
new cb9a458 Whatever bug used to be here before isn't here now.
new 61f3866 Use const reference parameter so it works with temporaries too.
new 94f640b Doxygen fix so that local paths aren't included in the built documentation.
new 8eaafe4 Actually catch all of the instances where single quotes start lines.
new d3c0652 Patch from Siddharth for #342.
new ca752b3 Document API change in an attempt to piss people off less when their programs are broken.
new c4d28a5 Clean up implementation of GetRecommendations().
new a2f5fd2 Update documentation due to API change.
new 32b0aa7 Spelling fix; thanks Barak!
new a4ef8fa Patch from Saheb for #301; refactor RangeSearch constructors so that leafSize is not a parameter.
new 5071776 Modify test for new RangeSearch API.
The 2119 revisions listed above as "new" are entirely new to this
repository and will be described in separate emails. The revisions
listed as "adds" were already present in the repository and have only
been added to this reference.
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/debian-science/packages/mlpack.git
More information about the debian-science-commits
mailing list