[shark] annotated tag upstream/v3.0.0+ds1 created (now bc5f699)
Ghislain Vaillant
ghisvail-guest at moszumanska.debian.org
Thu Nov 26 15:47:07 UTC 2015
This is an automated email from the git hooks/post-receive script.
ghisvail-guest pushed a change to annotated tag upstream/v3.0.0+ds1
in repository shark.
at bc5f699 (tag)
tagging 60332c7c30c7207a04d85f65465cda7ec9993d26 (commit)
tagged by Ghislain Antony Vaillant
on Mon Nov 23 13:04:57 2015 +0000
- Log -----------------------------------------------------------------
Upstream version v3.0.0+ds1
Bill (1):
mark boost as system lib
Christian Igel (2):
changed jiggle function
Merge branch 'master' of https://github.com/Shark-ML/Shark
Ghislain Antony Vaillant (1):
Remove upstream debianization.
Jan Kremer (7):
corrected typo
added permutation-based feature importance estimation to random forest and cart classifier
added oob error to forest and cart
added template arguments to NoisyErrorFunction
deleted IConfigurable references
removed previous changes
added workaround to make shark compile on latest mac os
Leninga (13):
doku
docu MarkovChain
RBM documentation
weitere RBMdoku
weitere RBMdoku
commit e841b7867ccb51b545cff97edb9e03a5badb13c3
commit 468bf38320780dbf89a6171bb5fa8f9d0e8577c9
fixed warning in analytics.h
Fussy: set m_parent in BellLT
corrected documemtation
corrected typos
adapted problems to batch mode and added doku
doku
Oswin (29):
making the code beatiful for github
updated documentation to reflect move to github
further updated documentation
Revert "making the code beatiful for github"
small indentation clean-up
replaced Statistics by something more useful
added new unified kernel for cholesky decomposition
fixed Cholesky issues due to bad ATLAS flag
added support of triangular_prod for marix argument
added set_element methods to classes that were missing them for interface unification
mall changes in documentation and readability
fixed issue with potr where the transposed call would lead to a compile error
fixed erroneous commit
fixed assertion alled in debug mode when vector was empty
fixed bug in packed storage computations. Not serious for production code
fixed boot linekr error on Windows and added proper warning flag on linux in debug
fixed copy&paste error
fixed OpenMP error in Windows as VC++ only supports ancient openMP versions
retyped DiscreteUniform to return std::size_t instead of long
code cleanup in RFTrainer.cpp by making variable definitions more local and fixing some type conversion errors
code cleanup in ROC.cpp by making variable definitions more local and fixing some type conversion errors
fixed a number of type conversion warnings on MSVC, more to come
clean up by repalcing blas::repeat(scalar,...) in sums by scalar
removed debug message
fixed indentation issues
replaced unsigned -> unsigned int in tests
Tobias found a bug in SteadyStateMOCMA. fixed.
fixed search & replace bugs
changed coinToss back to discrete
Oswin Krause (10):
added hacky float support for csv reading
Merge branch 'master' of ssh://github.com/Shark-ML/Shark
Merge pull request #2 from Shark-ML/blas
fixed mac linalg problem
fixed Statistics tutorial
Merge branch 'master' of https://github.com/Shark-ML/Shark
fixed some bugs
removed unused files
fixed two issues reported by christian
Merge pull request #3 from Shark-ML/vs12
aydind (45):
HDF5 fix. CMake will now test if the existing HDF5 lib has the correct executables installed.
Add missing flag for Clang 5.0 on OSX
update header template
update header template
add tool for replacing headers. will execute it this evening.
Update existing headers to LGPL.
update of header replace script
Remove all file tags
minor typo
add libsvm-like wrappers
remove small typo
remove wrappers.
Refactor matrix classes
fix for the includes. part 2/42
hotfix.. part 3/42
add a special kind of cache that just makes sense for kernelsgd
add vector include to avoid compile error on old mint 14
we need boost 1.48 at least
minor change (mainly for testing, so ignore)
add boolVector/Matrix..
make ACF and shrinking optional
allow non-const access to basis of kernelexpansion
rename PrecomputedMatrix, minor typos
add first version of kernelBudgetedSGDTrainer
fix errors and docs
add projection maintenenace strategy, missing any tests
update doc for projectbudget..
revert unintended checkin
add label order class
make valgrind and vs2013 happier than before
small doc fix
minor bugfix/update
make code more c++
add componentwise sin/cos
change test suite names
change test suite names, rest
add coverage/memcheck
update test cmake files
hide coverage, if mac was broken, this should fix mac build
fix stupid merge error. someone should ban me from using svn. or we move to git.
try to fix the stupid bug. its late, i need vacations
remove bogus and unnecessary file
update copyright date
add valgrind results folder
minor fixes in gridsearch tut
bjorn_weghenkel (9):
added an example for FisherLDA
skeleton for multi class svm approximation
changes for multi-class svm approximation
error function for McSvmApproximation
minor changes in FisherLDA
changes for McSvmApproximation
McSvmApproximation
messy code backup of mulit-class svm approximation (not working yet)
much cleaner code for multi-class svm approximation (still not working)
chivesman (11):
Import HDF5 file
comment out uncommited files for another task
*) make hdf5 optional and print skip message instead of warning in case of not found
Naive Bayes classifier
Fix Bitflip manipulation and thread.move in probe.cpp
include appropriate header files
use find_package for locating hdf5 package
commit the wrong file, fix...
(correct files)use find_package for locating hdf5 package
SVM supporting missing features
Fix Trainers_NBClassifier test failure
christian_igel (550):
commit 07b3037c8992dc06054b9894165c43f1592e8cd2
new dataset interface
bit fancier
for music paper
outputDimension = target.dim(1); needed for derived classes
for ESANN paper
new Dataset class, not tested
Why are the QWT people change their API more frequently than Felix Magath the team lineup?
new g++ smiles again
removed "false" parameter in resize after warning on Linux
Hi, corrected some minor typos ...
index shift
true positive rate added
commit 8931377a87e44cb60d1b318cdb7d618f7ebbfbe0
balanced sampling for multiple classes
balanced sampling for multiple classes
commit fcc1a11c5c70f835fb6332e9604a55de7751ac06
conversion to one hot
typos in doxygen
corrected typos
typo
typo
added amsmath
up
added amsfonts
doxy
typos
typo
typos
typos
typos
typos
up
up
up
typos
briefs
typo
brief
brief
corrected bug in documentation
doxygen typos
typo
added support of comments
typo
updated declaration
comment missing
redirect
stream
first version, no derivative
updates
boosttrap -> bootstrap
MS corrections
MS correct
MS corrections
typo
update for MSV
typos
diag problem exists
unsigned int
typo
importing PGM files
typos
initialization
base -> Base
works now under Windows
exportPGM works now with const arguments
changed Set to UnlabeledData - perhaps wrong
can now read a vector of doubles
virtual evalSet
fix of compile time error - please check
typo
2DO list for shark release candidate
updated style
added handling of of CMP0003 policy
initial, reviewed version
initail, reviewed version
some errors removed
compiles
does not work
significant progress, one main bug removed
reviewed
some improvements, support for log encoding
reviewed, test for area under the curve objective function
slightly better, gradients at least are correct
corrected random docu typos
docu layout
improved documentation
slowly getting there ...
reviewed
small update
stream operator added, reviewed
typo
new tutorial, some modification to existing tutorials
headings
added brief
typo
reviewed
reviewed
added knn
knn
typos, new links
Set -> Data
wrong infos corrected
int -> uint
evalSet -> eval
reviewed
reviewed
added fucntiosn to set parameters
added normalization option
undo change von () to eval
typo corrected
added PCA tutorial
first version of PCA tutorial
300 vectors
better reconstruction
Capitalization / .
reviewed, default is now release
reviewed
PCA test, reviewed
PCA, reviewed
small upadtes
made design matrix local
typos
typo
small revision
added references
typos
LDA tutorial
registered LDA
first lines of SVM tutorial
memo
reviewed
added SVM tutorial
first version of SVM tutorial
just editoril changes
adapted to new interface, works
corrected bug
reviewed
typos, added SVM special case
updated, information about 0-1 encoding added
reviewed
replaced by NegativeAUC
small edotorial changes, error corrections
reviewed
removed symmetric
typo
changed a \param
editorial changes
doxygen error messages
reviewed, shark_global
updated missing parameter description
doxy
removed image
up
missing entries
new tree frontend
generic update of documentation
renamed
updated, works with revised interface
reviewed
reviewed
2nd SVM Tutorial, not ready yet
StorageAdaptors.h reviewed, statistics probelm child
removed again
some updates
update
updated header file for trees
added global documentation
removed absolute path to Boost
smal limp.
small reordering
moving, moving
moved tutorial code to examples
moveing
small update
update
update because of move of examples directory
typo
corrected paths
download to doxy
improved documentation
improved documentation
Brief -> brief
improved documentation
improved documentation
mirco update of documentation, needs improvement
renamed class list
improved documentation
small doc update
undo
editorial changes
capitalize
update doc
improved documentation
typo
doc error
doc error
doc error
up doc
doc
first update for competition
reviewed
brief doc
set structure
improved doc
improved documentation
some updates
samll updates, downloads filled with stub
NonEmptyComponent
brief
improved documentation
better doku
made gradient test less conservative
draft of download page
slightly improved documentation
prelimenary
changed data path
added not to add boost libs
added -lboost_system -lboost_filesystem -lboost_program_options
added Video, updated FAQ
changed layout of citation
new box, 3rd heading smaller
small changes in format
rm .
small fixes
new box
all includes modified
added new subdirectory
test move
moving the rest
font change
Test prefixes all plural
removed some of the 2Dos
increased font again, reverted my own change
corrected paths, added numberOfClasses etc.
added conveniece functions, corrected typos
updated link to development plan
talked about location of exampless
typos corrected
rst style improved
typo corrected
minor changes, first try to fix math fonts, does not work
commented out math fonts hack
added new package
changed order
idoxy typos
typos
typos, added transpose signs
updated documentation
added @
idded doc line
typos
doxygen
update for doxygen
docs, corrected pot bug
correcte link t 64 bit Linux
small changes
tiny change
removed Fuzzy example
update before submission
catches exceptions for missing data files
updated path
changed path and file inclusion
new path concept
small update
moved spaces
missing training
updated to most recent version of AUC
up, AUC is back
removed explicit static linking
improved handling of dynamic library switch
switch of make test for dyn. lib.
data for clustering tutorial
reviewed
auto resize model
small update
added tutorial code
add unsupervised data
up
clustering tutorial
small updates
registered kmeans tutorial
added prize
prize, yeah!
update
as reviewed
review, bug corrected
removed, now there is a trainer
now there is a trainer
up
up
added Fuzzy Prefix
small updates
update typo in ffnet and update news
lib SVM
new ()
added blank
changed rest handling
initial version
small typos
explicit cast
explicit cast
size_t
added new qt flag
corrected label output
up
highestIndex option
typo
up
added convenience function
potetial bu corrected, was nict checked
added convenience functions
removed unused variable
up
up
added hidden, secret macro
added convenience functions for STL vectors
tree-based classification in Shark
sqrt -> std::sqrt
moved RF/CART modesl to Trees subdirectory
typos
updated reference
up
up
up
up
up
added set.h
up
up
requiring 1.45 again
updated remarks converning boost version
loading regresssion data
rm debug code
bug in cvs export
up
small fixes to documentation
up
up
up
up
up
up
up
up
up:wq
up
up
convenience function for initialization
up
up
up
up
added hidden threshold feature
up
up
output compiler
CI first reading
up
up
up
up
up
revised, CI
up
up
up
removed unused typedef
small updates
up
up
small revision of tutorial text
up
up
stray characters removed
https -> svn
fixed bug in CART trainer, I asked Kasper to check it
changed https to svn
up
tiny revision
link corrected
bug corrected, derivative threshhold was missing an entry for the regularization parameter
adjusted for new datset interface (hopefully correctly)
small improvements of documentation
typos corrected
up
: -> ::
added softMembership to LinearClassifier
typos corrected
made a function "const"
up
added svn protocol
rm errors/warnings under Ubuntu
corrected \file
updated bounds
editorial changes to the excellent tutorials
editorial changes
up
removed some warnings under Ubuntu / init order, size_t
test writes number of iterations
up
distance -> distanceSqr
must be tested, documented, etc.
was double
repaired
corrected wrong initial ifndef
deafault name added
up
added AE tutorial, still needs review
up
small update of tutorial
size_t
made output directory same as directory of the binary
revision, e.g., added transformation functions
typos corrected, bug #70 fixed
typo in parameter description
up
fixed bug in hello world tutorial
typo
updated cross-validation
small fixes, mostly releated to documentation
filename as option
some fixes, started working on tutorials (little progress)
doxy
fixed link to general optimization task tutorial
updates of the tutorials, work in progress
updates, read files via command line, prepared for new tutorial system
typo
typos
typo
up
atlas headers
mark_as_advanced
typo in comment
added boost dirs to top level
commented message out
// we should move to Boost.Signals2
added /opt/local hint for ATLAS
update
filename changed
removed probing
removed probing, added functions to get eiegndecomposition etc. to CMA
minimal update of CMA tutorial, which needs revisionm; CMAProbe -> CMAPlot, has been revised
removed boost regex
revised tutorial
removed bu just introduced
fixes small documentation things
up
warning in doc
added a stub on tut and tpp
updated description and interface to regularization network trainer
typo, but still does not work
bug removed
added note on multi-threded libs
made "official releae" advanved
test off by default
up
no test fails
no test fails
up
typo
doc typos
small changes regarding typs and casts
changed attribute table
up
bad typo
up
typo
up
up
small changes in documentation
changed JaakkolaHeuristic API such that original and faster algorithm is default
changed order of include files
up
added initial random number tutorial
up
isnan -> boost
changes constructor such that test work again
fixed probelm when uni / discrete is called with min=max; changed file encoding
added comment, removed useless argument from distribution constructor
MultiVariateNormalDistribution API updated
changed license
up
added news about LGPL
up
up
up
adapted for OPenSuSE
add comment on shared libs
updated
up, typos
Visual bug
added attribute counter
typos
added support for accessing columns / features in data sets
up
removed empty dirs
up
minimal test of default variant
removed bug in computing the angle of the ellipses
typo
insert function to set sigma
changed to unix format
1st attempt to fixe scaling
added counter
unused variable
updated CMA-ES to match the 2011 tutorial version
up
changed separator form string to char
small fixes for documentation
ref to RBFN
added missing kernel matrix includes
up
up
up
small updates
improved documentation
typo documentation
randomly swap loci
mac rpath policy
test for version
rm warning because of double (())
added documentation
updated doc
capitalization
up
up
ignore
small revisions
test, added loval js conf file
up
typo in documentation
bug corrected
newlines
small -> tiny
removed old csv interface, killed warning in IHR?
removed typename
removed references to Quickprop and OnePlusOneES
increase tolerance from .01 to .025
added "return solution;" after main for loop
updated proposeStartingPoint
propose starting point corrected
removed configure(PropertyTree const& )
up
typos
minor typos corrected
Oswin fixed bug related to vectors with zero length
Pole balancing RL benchmarks
Small updates of the documentation
didacr (2):
Fixed warning about signed-unsigned int comparison
Fixed some bugs and added some small feature for title lines in Csv.h
glasmtbl (356):
commit 1b5cb504cf5351e5f506b507b1f99ee60004b010
new solver structure for the Crammer&Singer SVM
new multi-class SVMs up and running!
ASSERT macro fixed (puh!)
Various Dataset constructors converted into Create... methods
Dataset syntax updated in all ReClaM tutorials
Fixed one of the ugliest bugs ever, of course,
commit 69be8c6b3a339cb72b80f07b125e7c21c6935398
commit 41d522e71f0a6ed5f6a1fe3840618b34d8b39c28
bug fixed in QuadraticProgram - finally!
commit 77ec8a4ba50725106f0a2d0a5cf32669cd2ce54c
commit 0ca3968416bb545a506a3f51a92ad0c9ce648ed8
commit 6fda886f866856595326d3baf09db0e8d292d3aa
commit f7405b1bf46d62374978833cc8efa0bab87bc47d
more QP solvers
commit 91520b75705ed1fc141bbe491883ab55b90c7424
commit cd541c1de3ee6c31d9e166d2d71c2e165ceadbdc
SVMregression example repaired
AbstractTrainer added
commit 8992f7395e6a8d16fa220cbe546f31d136d275ad
commit a1905ecdd09ffc0a5ee12c1edfb85be6296b7caa
McSVMs mostly done
commit c3e2935df50d407a926e2368ea0cada9094bc9c5
LOTS of changes ;)
it compiles! :)
commit aa4f614341bd4af837de083423839704d1a9f2c2
commit 2b34ad534d9203c9c1a4a1721cabaaebea110a64
Converter vergessen ;(
multi-class SVMs are working
commit 2a1cbf12e50ce821e5c70fa2ad2620635a864b66
two new McSvm machines, and bugs in the QP solvers fixed
one-versus-all multi-class SVM added
kernel regression ready to use, solvers split into multiple files
Dataset added
commit 778f210e2056ac724cb290793d015ff9ed7166b8
commit 055bd643db9830a4eb632f5b6d3ec239694ac560
commit b7dbf673f49d000739f604a8d5c9f9f698b84d2d
DataDistribution added
stupid solver bug fixed
commit 7cdc35abb8c216234b86ff629931aa8024bfa924
commit de9096caf28dceabb7bfeb516401f7e8cbca4007
normalization for datasets added
dataset changed to text format because boost::serialize sucks...
inputSize and outputSize removed
commit f588fe3d3bda955dbfcad7fe1bf53f27bfc4b4ef
commit 940375a50d2b8e62130cac58329e1cd262add4b5
SVM training is now based on templates instead of inheritance.
commit 885e627a95f454bc7c76c0f13162d99a489071e4
commit 1b5bfea21dd4945e7c0af6802eef692625c0b642
commit 10b01c7a44a82c31fb9b6744906b1b29543c19bb
stupid bug in Dataset fixed
kernel regression working and tested
commit 53cbbaca269f2f99ad9cf667fb4848a46097f6b7
commit 5257bc8c5fc748a3f3ce12e55d815144fb8aaa4f
commit 111d4813b3e886109f6a685ab0b1162377d4ad5c
commit 85b7d358ceeccd9cf790467f1196ce0c718da397
commit 38d34e131214677d6e70fbd34acabd5ef8ef6709
numerical problems in QP solver fixed
numerical issues in QP solvers fixed
bugfix die Zweite
sparse vector and matrix types and shark::Slice added
commit d8b8af1c724587db8221ed3bf3af917bdc56027e
dataset const_iterators added
boost::iterators for Set and Dataset, and CSV import/export
GLPK simplex solver added
LP test case added
completely new multi-class SVM solver
commit 70b6cf8f0718cf6af586ffbe5e6423e2edfd896a
reorganization of the directory structure - I hope everything is still there...
commit be2f5565759de8a128fb16db8193a3e26a36284e
commit d4b51ad59f62c059ad0e0c021a782fd486c40b73
commit 7e62b28e1f95da177215a7b8289e1abbee34af71
commit 41a358175e3e34c8fca2c4bd0b0cf0aed578d9a8
template parameters added to LinearModel and normalization trainers
corrections for interference with Oswin's checkin
s2do versus smo switch added
multi-class SVMs renamed according to unified naming scheme
SigmoidFit test case added
NormalizedKernel derivatives and documentation fixed
commit ff8cd64bb4da3e6c5d23e42d72106fc93f9fb34c
IParameterizable interface added, cross-validation re-designed
absolute loss and test cases for the loss, and for cross-validation added
test case for CSvm training added
commit f66edf1c2ad4666c5910458774a87575f2e75c43
commit 7cefa64b157f90e63462492d4e04b1a541a75636
xerror changed, ellipse avoided
commit 2b539ab9df468f94ad16b06b784b19ba6edbb946
a big bag of fixes
many macros and headers fixed
commit 6443de3dc41829e86fe23eeda6949d6e2366051b
evidence, derivative ported(?)
my first attempt on the documentation :)
evidence updated, diagonal matrix type added
documentation on optimization and trainers
coding convention moved to sphinx, and slightly modified
conf.py changed to relative paths, link-to-doxygen tutorial updated
ToDo list created
commit fc98c9f3c5cc941dbb617623843504c4f525483f
Set and Dataset renamed into LabeledData and UnlabeledData, respectively.
ARD kernel reviewed, minimal (editorial) changes.
sqr -> sqrt (bug introduced in the review, now fixed!)
Main change:
commit 594d35c0137705284ad1e219bedb608d21575b02
added data tutorial, worked over old data tutorials
attempt to fix the double -> float conversion warnings
AbstractCost added
quick bug fixes
commit da1d0f309693ddeefe4e43b20163b438b8f187c1
iterative nearest neighbor queries
TODOs added
fast nearest neighbor classifier based on binary trees
commit 9aff86c5e4853cc5aba0cb80f630fa6c80d87d4d
kernel expansion with unit test
KTA test case
kernel functions split into multiple files
nearest neighbor regression test case
const references in k-NN models replaced with pointers,
commit 23fce5b3704b6e1634f43c1cdc8faf1af4fa5b55
bug fixes
commit 7f2056e41a7fb1df58edf1038f6251ee9412823d
unit test altered, hope it helps :)
still annoying unit test debugging with 32bit Windows
file headers added/fixed
commit d92be5482c50c637279c14aadbc2396fe57e7850
commit f3c12d5ab8a90e14bc1dd010f9d8497beebb3f09
editorial changes, see review request 76
check in of review request #84
documentation added
Christian's comment on documentation style incorporated
check in of review request #86: interfaces for clustering
commit ad917a1e9bd79797e429ae25550b7e4222f20963
check in of review request #102:
oops, forgot one file :)
and another one :(
check in of review request #103
check-in of tree-rework review request #96, with comments incorporated
location of trees changed in the directory tree
unit test for soft nearest neighbor classifier
include macros fixed
check in of review request #134: "McSvmTrainer fixed"
bug fix (index error caused crash on windows)
more meaningful grid for the example program
"\brief" documentation added
check-in of review request #137, "new readme for Shark 3"
commit 6b03e5ecf4e1243fba6e0e8c38c9abef39c5b57d
check-in of post-review #146: bug fix in binary space-partitioning
little code clean-up, removed unused old stuff
macro names updated
checkin of review request #147: k-means
documentation improved
derivatives of the weighted sum kernel fixed
added derivative of radius margin quotient
leave-one-out error added
Leave-one-out error for C-SVMs, as well as a few related changes to QP solvers.
ooops, forgot the main file :)
Nearest neighbor bug fix. This slows down tree construction from
Polynomial kernel added,
k-NN bug fix - I really hope it works this time!
unit test improved, still passes :)
documentation extended
small bug fixes
fixed KernelFunction unit test by removing a bug in NormalizedKernel
bug fix for Pengfei's numerical issue
This is the first promising attempt to fix the long-lasting
unsigned int replaced with std::size_t
bug fixed, thanks to Xuan Tuan
fixed unit tests for ancient apple gcc
small improvements of the documentation
"typename" removed (gives errors on "ancient" apple gcc)
bug fix: initial shrinking is not called only if shrinking is enabled
First version of a Pegasos solver for linear multi-class SVMs added
LinearModel changed from mapped to compressed vectors/matrices,
warnings (hopefully) fixed
fix!
objective-function-based stopping criterion removed from SVM solvers
another bug fix, thanks to Xuan Tuan!
one-versus-one classifier model
made the import more flexible; classification labels are automatically translated into Shark's format.
commit 152fdd96e8fae8b37d597b1a74c4681f9b27b535
Libsvm import problem fixed; unit test extended.
small bug fix.
another bug fix thanks to Xuan Tuan
convenience constructor added
linked boost libraries to the Version executable; now it links successfully with boost 1.5.0 and gcc 4.6 on the current Ubuntu.
trim added to csv import
OptimizationTrainer adapted to recent interface changes
sorry, loops fixed.
random forest changed in
documentation improved, name added to CART model.
Check-in of review request 208 (fixed clang warnings).
Check-in of review request 209: added product of kernels and a kernel wrapper that applies a transformation to the inputs.
addendum for review request 209: changed cmake files
commit of review request 210: new tutorial
Check-in of review request 212; addition of a special kernel for multi-task and transfer learning.
documentation added
three new unit tests added,
jaakkola test moved down in the list
SVM trainers split into a multitude of different files.
forgot the most important file :)
SVM documentation greatly improved
documentation improved
Clean up of the solver: old debugging stuff has been removed,
small updates on the docs
more text added to the tutorial on why derivatives are the way they are
more minor style corrections.
serialization tutorial added based on Lauge Sørenson's input
several 'e's added, and other minor language corrections :)
commit 5ec49f1ae35ef48382679d50f282efd0d2fd3c5d
changes in comments only
liblinear style training for linear SVMs, PEGASOS deprecated.
commit ca72b816ecd15dd9d7c9bc30cca052bc0b2e8f61
commit 4531956c6632ac41c55c7399ed492d0cc3d64831
temporary debug code removed (sorry!)
superfluous "typename" keywords removed
changes in spaces, as well as two tiny fixes of outdated documentation
default value of second template argument changed
This is basically a revert to revision 908,
linear SVM training changed (follow up on yesterday's change in LinearModel default tempalte parameters);
CMakeLists.txt fixed (hopefully!)
Data normalization changed.
commit 76d3c459e7a6ed44e40a4243bd88b764dfc26312
Ugly bug in OVA SVM training with CV fixed,
made destructor virtual (bug found by Sebastian Houben)
New unit test for linear SVMs added,
bug fixed in ATS and ATM trainers,
forgot one :)
bug in linear SVM solver fixed, THANKS, OSWIN!!!
Data distribution split into supervised and unsupervised cases,
forgot one file, sorry!
better parameter check
comments only
Network component moved and replaced. This is still a stub,
just another stub
web server component moved (again)
Web server component up and working (at least on MacOS).
small fix for clang 3.1
Oswin's find script
comments updated
new tutorial on how to write a kernel function
webserver component tested on Windows
preliminary version of LASSO regression,
LASSO regression, now fully operational.
experimentation with failing unit test
LOTS of changes, due to
changes in file headers only
nothing serious :)
tree nearest neighbors bug fixed
kernel regression unit test improved and split into
imcomplete xNES implementation removed
fix provided by Craig Brown
changes in comments only
changes mostly in comments only;
quick fix for multi-class trainers; disabled shrinking for "max-loss" machines.
The netlib dependendy should now be obsolete.
added some preliminary documentation
generic "archive" wrapper for objective functions added
fixed and reactivated the TSP example program
First step towards including example code in tutorials.
typo fixed
command added for including whole source files
web server tutorial stub added
syntax error fixed
multi-task learning example added, still buggy
some preliminary bug fixed, need to convert to MultiTaskKernel to batch interface
commit d11adc54265c78eb5eeaf5865b2b135049b623cd
Input type of linear SVMs has become a template, allowing for arbitrary vector formats.
some convenience functions added,
Complete restructuring of the web-server build system.
web-server docs removed from core library
small bug fix
sorry, forgot to check this in earlier
headers adjusted,
first version of kernel logistic regression added
training with offset enabled, number of epochs adjustable
documentation improved
bug fixes for KernelExpansion (m_outputs removed) and SGD (epochs initialized)
features are now allowed to start at zero
better (?) constructors
KernelSGDTrainer not inherits IParameterizable
minor typos fixed in the docs
BUG FIX: libsvm import now throws if file opening fails.
comment header only
Archive wrapper for objective functions added.
Constraint handling bug fixed,
span bound removed after debate on the mailing list
inner iteration limit introduced, solves some (numerical?) problems
ooops, sorry, now it works
header cleanups
"five" changed into "six"
Fix the Dataset crash with clang on the Mac,
attempt 1 at solving these f**king cmake issues
GLPK error handling fixed
version 0 of lasso toturial added
commit 3d825609894d4c5746a00f42f725d47a9fe95bdf
commit 60572eff0af19a4f4eed3e9654a6d7102f4de1fb
Versatility of learning tutorial fixed, all code moved to example programs.
Worked over statistics tutorial and example program.
commit caacf0169733e668b3ae8e18066ae1c8c8a5d734
expression -> expression() bug fix,
Oswin's noalias fix ported from the LinAlg branch to the trunk for immediate benefits
commit 02f6ab68dd053802c5cdda11f80952a9170d2aec
simple "make uninstall" added
timer problem fixed
two new tutorials added
examples for tutorials modified/added
header
bugs fixed
more fixes
fix of ugly "& &" --> "&&" bug
buggy (?!) third parameter removed in csv_import for labels
another timer fix
commit 1f62194208f01e260a7dc8bb0b8c9aaaeeb5b197
rudimentary png support added
clang warning fixed
commit 87006a4ad827788b6cd308f70cd3aac2e7628aad
macro renamed
GNU linear programming toolkit removed as a preparation for license change
commit cedddd57eac3aaa945f82cfa64c32e44b9156d1f
commit d6a3f762a7a8f79f7c5fe5fe3b21731e08af979f
"inline" added to remove clang warning
1/2 added - gives correct results but is somehow odd...
commit 3b7d604682d99876ae5f26335d0b7c2e821c2c1e
first checkin of reworked data tutorials
markup fixed
commit cdf53cd733e248511fdbc10d7bf9af0950be1ed1
more on data tutorials, this time with a bug :(
commit af60d2ca6e35e6b56eb856c310423b3ab0c59a60
data import/export routines renamed, examples + tests + documentation updated
commit d05f97146425e6a115f20bb5f664413bf5f2aa4e
forgot one file
commit f3cacf5b8bd2d2fb67fdbebab394f902306f1ed5
Oswin's comments incorporated
commit 92553631db244b062815896909aa22f3b4b8b6bd
(too stupid to proof read...)
documentation slightly altered
commit e3c2627211d133bcc56fd38587c70a71ae9f3ad0
"bug fixed" - or rather unit test adapted :/
Added a trainer for the so-called reinforced multi-class SVM.
unit test and example added for reinforced SVM
1/distance weights added to nearest neighbor models
missing includes added
linear reinforced multi-class SVM (RM-SVM) added
additional constructor the soft nearest neighbor classifier
Added a clearModels method to MeanClassifier, which is now called by RFTrainer::train.
but now finally...
solveSymmSystem renamed into solveSymmPosDefSystem
prod(RealMatrix, RealDiagonalMatrix) enabled
removed default of constructor parameter, added more documentation
documentation updated
#include <set> added
Flags for external linear algebra functions reorganized, preliminary support for the MacOS Accelerate framework added.
simplex downhill added (not tested yet, will try to do that tomorrow)
Simplex Downhill bugs removed and simplistic unit test added
typos fixed
"real" unit test added
unused variable removed
kaibruegge (3):
changed template parameter so Contrastive Divergence can use different transition operators now
Added a configure method to the transition operator
added missing include statement
leninga (38):
documation errrors behoben
restliche Dokumentationsfehler behoben
still more RBM-doku
doc RBM/analytics.h
commit 94ecfbb28fb3f916938d0d14b0bff7e7daf1a1d5
docu again
removed function setMatrix()
commit 440946fa44d3c73480f4e3db19336762db163577
new configure()
commit 83444299ba4c0766e80c446841e4553d73d26c97
likelyhood -> likelihood
commit e45700b8865686f370e2ab854f1ecc45935ec213
added doku
doku
comments, corrected names of the output variable
added weightSum()
corrected rule.print()
added function Rule.addPremise()
Fuzzy: set m_parent in constructor of all LTs
commit 8cd17f9fd9544c60c19db091837667f5cad7cc30
RBM-Doku
error corrected and RBM-doku
RBM-doku
RBM-DOKU
RBM-DOKU
doku
typos
corrected size() and element() for batch
corrected size() and element() for batch
commit a56a5ef22217c6cfef4cb3a3d2be4c54d7f0791f
RBM-Doki
RBM-Doku
corrected docu rangedSubset(...)
typo
removed resize()
RBM Doku
RBM doku
RBM doku
m-p-t (446):
Added alternative ctor that can consider only a subset of columns of the input data
Added alternative ctor that can consider only a subset of columns of the input data. re-committed b/c of new repo path.
current libsvm versions also support multi-class files: small update to libsvm saver
also update libsvm saver from 0-based to 1-based class and feature indices
added loadLIBSVM, and updated SaveLIBSVM for consistency
minor adaptations to new constructor layout, and 2 small label conversion utilities
tiny adjustments, for overall consistency of dataset handling
fixed a tiny bug in libsvm-reader
beta version of epoch-based crammer-singer solver
also add a flush after each verbosity-2 output
- split all solvers into sub-files: QuadraticProgram.h now only includes all solver headers
forgot to remove now-oblivious QuadraticProgram.cpp
added documentation
commit d3512c49756cd515aadcf7c620d73bd43c83e00c
templatized KernelFunctions in KernelFunction.h,
implemented ci's suggestion: remove stochastic component in cicl-target-noof-hits determination
added alternative candidates for sparse vectors in preparation for some benchmarking. not going down to the level of un- or bounded_array sparse template parameters yet, however.
- added an import method for LIBSVM-formatted files (export and import-to-set-only are still missing)
forgot to remove some debugging lines
fixed boost split bugs and refined libsvm import
slight changes to distanceSqr
- again optimized distanceSqr some more
- added some overdue ignores
a second attempt in win-compatibility
milestone preparation check-in
finalized the wall-clock timer cross-platform issue
initial commit of sphinx doc folder
change to gplv3. please include according headers to all files currently without one.
small fix for documentation parser
small fix for documentation parser
fixed my editor going whitespace-rampage
added ignores for example executables now generated in new place
shaped up oswin's initial commit and took the opportunity to rework the rst source tree,
bugfix in relations sidebar and some small rearrangements
1 -> old tiny change for corrupt libsvm format as e.g. present in lasvm data
added some ignores
- incorporated jan's copy constructor and = operator
added some ignores
- moved sphinx output to its own html/build/ directory
re-organized sphinx tutorials
- re-organized sphinx tutorials
re-organized the home page:
- added faq
doxy-sphinx integration
preparation 1
preparation 2
preparation 3
only document source folders -- plus, there might be other configuration changes we might want to do here!! i recommend a diff to the shark2 equivalent!
preparation 4
preparation 5
added file suffix patterns
great! -- truly relative pathnames
doxylink is now shipped with shark and called with a fixed relative file name. simply from contrib/doxylink/sources run
fixed ignores so build/lib/python may actually exist but no contents seen...
added missing ignore
commented out the doxylink part for now.
minor updates to documentation system
added a few ignores
comment out tag file by default
cleaned up docs a little
cleaned up docs a little
ADD:
ignores
renamed weighted... to diagonalMahalanobis..
small bug fix for derivs for negative ARD kernel params
- fixed a bug in Libsvm reader (please tell me if you change the label interpretation! :)
forgot to add the test...
getting rid of some near-superfluous cosmetic edits i want out of the way when trying the review tool.
[AbstractObjectiveFunction] FIX: boldly fixed the html-doc-generation bug. not pretty, but works at least
[AbstractCost] Typo
[KernelExpansion] Fixed bug in docs
[AbstractObjectiveFunction] Had deleted too much..
[NCLL] docs
[ErrorFunction] added missing include
[datasets.rst] small bug and language in datasets.rst
[datasets.rst] another small bug, and language in dataset_subsets.rst
[data.rst] added a 2nd warning on +-1 vs 01 earlier on.
[tutorials.rst] no need to prepend paths when referencing a file (i.e., that doesn't work)
[tutorials.rst] no need to prepend paths when referencing a file (i.e., that doesn't work)
[concepts/overview.rst] registered the cmake tutorial
added missing headers to two supervisedObjFcts.
fixed two small bugs
small typo
small bugfix
some changes to the labelConventions -> need further attention/cleanup
cosmetic changes
bugfix: forgotten include of iostream
several bugfixes and new example on how to link to and static text files
oops, forgot these..
and a bit more clear
added ci's new tutorial
- evalSet -> eval
for some reason, sphinx doesnt like the same reference over multiple files...
clarified section on thresholdConverter
useless cleanups
you do *not* want to know how long this took me
small typos
small bugfix
updated SigmoidModel, SigmoidFit. Incorporated several suggestions by tobias into sigmoidModel, namely on encoding and input types. Added an rprop-based wrapper for sigmoid optimization to SigmoidFit. Wrote a solid test case for the latter.
small comment on the state of the platt unit test
fixed small bug (samaria, not samarian)
improved documentation
small but in documentation
added inputDerivative and adapted test accordingly. Also removed weightedParamDerivative because the DerivativeModel version should just be as good when there is only one output, i.e. no computational gain achievable.
whoopsi
several rearrangements and small improvements, also added a reminder-skeleton for the derivative tutorial
tiny: mentioned test cases as source of information
sorry, can't :doxy: a RealVector (yet?)
added shark_globals group, reachable under
celebrity bug
orphaned AUC.h broke the build
fixed small doc-bug in invert.inl
small bug in css active class recognition
added explanation of how to install doxylink and sphinx
some spell/grammar/other fixes (hopefully not corrupting the meaning)
small typo
small typo
replaced pngmath by mathjax -- works for me
small bugfix for McSvm operator=, found and requested by the RTCV group
several typos/bugfixes/updates. reverted mathjax again b/c of personal repo corruption. can add later again if we want it.
several typos/bugfixes/updates. reverted mathjax again b/c of personal repo corruption. can add later again if we want it.
several typos/bugfixes/updates. reverted mathjax again b/c of personal repo corruption. can add later again if we want it.
tiny restructuring
adapted sigmoidFit test case to new label convention
TODO.org is replaced by the trac
moved all kernel stuff where it belongs, cleaned up docs
correct small formatting warning
small doc bug fix
major documentation overhaul according to cis comments -- i hope everything works :)
migrate documentation/tutorials to tutorials
forgot one
somehow the post-review -r thing didn't work, so here goes. if you don't like something, change it yourself :)
deleted orphaned doc dir
wiped getting help again (sorry :). it now lives in the faq, feel free to move within existing pages
more documentation work
minor reorga, and added tvs statistics tutorial
two small typos/styles
clean slate for the sum of outdated license, install, and readme files. consciously re-write from scratch if you miss something.
incorporated cis latest comments
moved some things around
removing two files now generated by CMake. next will try to add them to ignore
alright, apparently that was stupid..
some minor changes, hopefully improving the file
another minor fix
- minor update to README: added info, copyright info, installation reference
removed two lines of (hopefully) meaningless code
re-introduced *.cpp to the doxygen file pattern list (for doxy example linking)
updated docs:
sorry for the doc-break, i'm again trying to get rid of these build artifacts
ignore shark.dox and conf.py build artifacts -- i think it finally worked!
updates for the new graphviz/dot dependency for inheritance and collaboration graphs
little touch-up
fixed typos and broken links
reverted to mathjax=NO
set math font size to 12. not ideally aligned now, but readable again.
updated documentation for competition
updated documentation for competition
made doxylinks relative (hopefully)
fixed one warning and a few typos
one more
fixed a missing latex def. please also run make doc to check that you didn't produce latex errors.
tiny changes:
fixed duplicate reference abbreviation
svn ignore
svn ignore
CSvm derivative now also works for no-free-SV solutions
updated KernelExpansion and its test.
fixed RBF test.
added hint from mailing list to win installation instructions
no idea how i can see which svn props actually changed.. well, here's the commit..
incorporation of log encoding into code and test
- finalized C-SVM derivative, including tests for C and kernel params
re-introduced some (top-level) template default parameters
- added test for log-encoding of kernel parameters
finalized test for CSVM-derivative. all done now.
final-final
fixed three bugs in sigmoidModel :/
tiny documentation things
typo
sorry for cosmetics, but my editor refused..
corrected documentation
tiny comments
bug fix
tiny doc
improvement for doc installation
mlms model selection
tiny, in preparation for update
svn ignore in preparation for python3 update
svn ignore in preparation for python3 update
update for doxylink 1.2
delete of test files for doxylink 1.2
delete of test files for doxylink 1.2
updated documentation system docs
duplicate version of KOSSA project plan
corrected path to KOSSA doc
had to disable pngmath_use_preview because that suddenly resulted in weird errors.
fixed last obstacle in KernelExpansion LogEncoding test (adapted solver accuracy for 32bit systems. strange error, but now it works..)
removed legacy non-shark assert definition
fixed a bug in sigmoidModel (reverse-encoding of underflow log-values)
renamed log_encoding variable names to "unconstrained"
code status update
added outstream operator for LabeledData
added note on non-portability of serialized objects
added note on non-portability of serialized objects
added one more sanity test to KernelExpansion test
svnignore
added new dir for test input/output files
removed obsolete file
use separate data in/out dir for testing
svnignore
added optional sanity check for negative labels
improved docs
changed order in which createCVIndexed works from reverse to regular
bugfix
svnignore
grammar bug
added data file for upcoming SvmNCLLFittability test
tiny doc improvement
clean up debug debris
whoops..
pami obj-fct up and running
trying to fix the jenkins-errors
trying to fix the jenkins-errors
changed last import_csv test to in-code data. deleted old dataset
small fix
added to documentation, in order to save anyone half an hour who needs to access solution()
clarify exception message
improved tests for cross-usability with SvmNCLL tests
(hopefully) last bug corrected, tests greatly improved
exp -> std::exp
avoid compile errors in default build mode (this took me a looong while..)
commit 14143ff1f80d7f5968b417a707184ff57bc0c922
reverted Test/test_data directory to avoid strange path errors on 32bit and win machines
improved docs on building the docs
added tutorial on how to prepare/carry out a release
added todo to clarify hdf5 status in documentation
docs
grammar bug
docs
remove obsolete bogus
remove unneeded header
preparation for new tests
preparation for new tests
improved docs
svnignore
completed splitfile code and test
added missing const qualifier
Jenkins has troubles with new paths, for some reason.
bugs..
typo
small bug
added opt-out for scientific format
added pami-toy-distribution
added pami toy test
added new data distribution: positive circle in negative square
added column-aligned output of csv files
doc bug, re-organized svm tutorials a little
follow-up
whoops..
commit a8ad364d2a603d68d1c1a8c3839c94f24c5e74a9
very sorry..
improved docs
bugfix
docs
meanvar for matrix input types
removed artifacts
added pami tutorial
finalized pami tutorial
split installation doc, improved docs
clarified atlas reqs
some docs
small docs
small docs
fix for compiler warnings
fixed "my" compiler warnings
- attempt to fix the MS2010 compiler error.
another compiler warning
another compiler warning
-typo
reverted to old lamda variant (the jenkins servers didn't know phoenix..)
tiny docs
- added weightedSumKernel1 from Shark2. weightedSumKernel2 is now FullyWeightedSumKernel.
- added macros to get a vector subrange type from a vector type
- added const ranges to vector range macros
- reverted const macros (misunderstanding)
- added, unified, and renamed macros
incorporated tobias const_cast replacement
- moved CSvmDerivative from KernelExpansion to its own file
- removed comment landmarks
a few fixes
- added scaled kernel and test for it
trying to make jenkins happy..
trying to make jenkins happy..
final add of kernel counter, some defines and comments in the solvers, fixing some shrinkcounter things
- added new toy data distribution
catch a popular mistake
- remove g_inverse test
small note on serialization from oswins mail
tiny but usually helpful edit in error msgs
slightly improved circle-in-square toy problem
tiny
fix doc bug
spelling
bug in doc
two tiny edits
tiny output format improvement for exceptions
fix two more bugs when reading libsvm files
svn:ignore
bug fix
eliminate a compiler warning
eliminate a compiler warning
svnignore..
added export functionality for precomputed matrices.
added one more export option to precomp. matrix export
fixed small bug when using sparse types
incorporated Bill's recommendations on post-review documentation
fixed csv read and write for sparse data
svn-ignore
added start() and stop() functions to timer which ensure monotonicity
somehow, sphinx doesnt honor the scale directive to rest-images
tiny documentation edit
tiny: added scope for sqrt in LCTree
cleaning up: store test intermediate output files in own folder. did not work on jenkins somehow, but since he is down we might as well tidy up
svnignore
read in csv and libsvm files irrespective of the line ending encoding
svnignore
fixed test output dir creation for out-of-source-builds
documentation bug
improved docs on labels a little bit
adapt to upstream changes in doxygen version 1.8.1.2
fixed documentation error for shark default file names in dataset load/save
- add missing line ending to faithful.csv
svnignore
extended label sanity checks for read-in routines.
forgot to update the sphinx css file as well..
svnignore
light overhaul of the installation instructions:
avoid inline code in headings (alternatively, one might style it better in the css files)
small correction
followed ci's request to rename SvmNCLLFittability to SvmLogisticInterpretation
- fixed compiler warning in glpk
incorporated awille's and DJ's mingw experiences
changed minimum CMake version to 2.8 for SVN in lists and docs
- tg: when adding a tutorial, one has to register it in index.rst
- added info on cache size setting and stopping criteria to svm tutorial
apparently missed one line when i did the shrink counter changes back in march
fixed libsvm test
update docs in preparation for ci's doc page update (still missing new SVN repository path)
several changes to the QP components:
- fix CMakeLists.txt to allow testing under dynamic linking
missed a file
svnignore
- improved greek language support
svnignore
updated docs wrt dynamic linking
- fixed bug in SigmoidModel weightedInputDerivative
removed legacy file LinAlg/Impl/Base.inl
conforming to a request by ci
new doxylink version 1.3
arghh
tiny docs
svnignore
tiny docs
upgrade to doxylink 1.3
upgrade to doxylink 1.3 - cleanup
upgrade to doxylink 1.3 - cleanup
tiny docs once more, sorry
- back-porting of doc changes from main branch
added tests for unconstrained version of rbf kernel, but strange behavior influencing the DenseScaledKernel_Test, so one is commented out
trivially fix kernel function test
whoooops
allow for non-continuous labels in csv and libsvm data files again
added test for bias-less sigmoids
improve CSS for sphinx pages
improve CSS for sphinx pages (backport from batch branch)
(also style content entries themselves)
(also style content entries themselves) (backport from batch)
more fully test the ard and rbf kernel encoding for unconstrained optimization
model tutorial overhaul and some css polishing
another css backport
final polishing / results of discussion w oswin
switching back to mathjax again
slight clean-up
slight clean-up of mc-svm docs
prepare local mathjax installation
svnignore
overhaul of kernels tutorial
svnignore
hopefully final word on mathjax..
added new folder for python update to 3.3
svnignore
removed new folder for python update to 3.3
docs now support python update to 3.3
prepare directory clean-up
directory clean-up I
directory clean-up II
some doc changes (in the works...)
more doc changes (in the works...)
more doc changes (only few more files in the works...)
done with this round of doc changes
further docs update
small quick update
* ported CSvmDerivative and SvmLogisticInterpretation to the batch branch
fixed a bug/inconsistency between the way the SvmLogisticInterpretation and CSvmDerivative on the one hand handled their hyperparameters, and the way they were handled in the CSvmTrainer on the other. updated docs and tutorials accordingly.
libsvm export should be sparse by default
- fixed bugs in data/libsvm
- tiny docs
a few small doc changes
- remove spurious output in BinaryTree.h
tiny docs
- small typos
removed doxylink as contributed software, now leaving its installation as responsibility to the user. updated docs accordingly. cleaned up conf.py.in.
forgot lines in CMakeLists.txt
- improved docs
svnignore for new tut2rst, and some small corrections to tutorial and sphinx config
* added more svn ignores
svn ignore
* correct small bug in sphinx markup
* svn ignore for in-doc doc builds
* removed obsolete fullyweightedsum kernel from docs (was removed from code earlier)
clean delete of mkl-ckf.rst in preparation for tut-ification
eliminate some sphinx/doxylink warnings
svn ignore for new mathjax version
fix bug in WeightedSumKernel.h (m_numParameters not initialized in ctor)
docs: update boost link to persistent url
attempt to reduce mathjax timeout when offline.. not sure if working properly, but not working worse, at least
minor typos
small doc improvements
svn ignore
updated mkl tutorial
renamed mkl tutorial
convert csvm max-likelihood model selection tutorial to new format
undo accidental ci of rst file
fix indentations, svn ignore
brought MklKernel test up to date. still commented out in Test/CMakeLists.txt , though
svn ignore,
fix (or evade via ignore) remaining doxygen warnings
svn ignore
- let Csv test also test comment functionality.
- svnignore
* fixed tooltip bogus content with new doxygen versions (>1.8.5)
- svnignore
- added better capabilities to recreate splits from indices: added index vector type to class signatures
- svn ignore
- expanded on chris recent changes to change the latex \vec command to print bold symbols instead of arrowed ones. In detail, add local mathjax config file, load it, and also load it for fallback local code. Plus all the same for doxygen pages, too.
- repaired previous hacks for vector redefinition in mathjax for strictly no absolute paths (e.g. allowing copy to diku server with docs built elsewhere). seems to work
oswinkrause (1140):
Added CMACMap
removed old temporary file which is not used anymore
LinAlg Tests portiert. CMakefiles angepasst. running
removed a debug-line in CMSA. added EALib Tests
templatizrd complete LinAlg library. So every occurence of Array, Matrix, Vector or an Function pointer was replaced by a template Argument
Complete transit to uBLAS in LinAlg. Working! Added testcases to previously untested algorithms. Working! Rest of Shark is broken due to this big change (except rng)
little fix to the signature of invert
I missed to update fft...
fixed a bug in CMACMap
fixed a bug in CMACMap
updated EALib and MOO-EALib. added recent changes of trunk. removed Array
I spent the last hours to fix the broken BFGS. Now it passes
fixed the derivative of AffineLinearFunction
fixed CG with Cubic Line Search, cleaned up the BFGS interface and added a missing file from previous checkins...
some more Tests and rewrote GridSearch
Added Tests for GridSearch Optimizers
changed a few more classes
Added a test branch for the new Interfaces. More to come
ersten Entwurf eingefügt, alte Bibliotheken entfernt
fixed everything...does compile, but not work (due to commenting out dlinmin)
Bugfixes am Interface
added a cast because of compile errors
fixed again derivatives of linear Models. Didn't work for more than 1 output
changed libname from Shark to shark
Updated AbstractModel and implemented Linear Models - currently untested
added shark namespace to LinAlg
added missing typedefs
removed protected function name() because of compile errors when calling std::cout<<foo.name() when foo is not const. use m_name instead.
changed layout of ReClaM to ML. Added CG and gixed the LineSearch issue by adding a new (pseudo)optimizer LineSearch in Optimizer/LineSearch.h. BFGS and CG are tested and both working.
adding tests and updated makefiles
implemented SquaredLoss and ErrorFunction (without 2nd derivative), test is working
previous commit crashed somehow...so version 640 is screwed, but 641 should be working again
tested and fixed the linear models
sometimes i hate svn
Ported the Rprop Family. Tested them via Rosenbrock. working
implemented and tested Quickprop/SteepestDescent
implemented GridSearch. Also changed the buildfile such that it enabled the DEBUG-Define in Debugmode
I again hate svn
fixed serialization and changed the interface ISerialization a bit, so that it supports the boost magic
fixed error in CG, sorry for that
implementation of network - looks quite messy
implemented the Sigmoid Layers
added FFNetwork-Family. compiles but currently untested
changed DataDrivenOptimizer
Added CMAC and fixed FFNet. working
added optimizer and NoisyErrorFunction as replacement for StochasticGradientDescent. added shark namespace to Rng
Fixed Compile Errors and implemented Kalman Filter
Implemented Softmax and SigmoidModel
implemented Model::generalDerivative and changed Model::generalDerivative such that it doesn't calls Model::eval() anymore
implemented generalDerivative for all Models
added generalDerivative to the CMAC which i forgot earlier...
fixed derivative of Softmax - the derivative should not be input(j)*t but input(i)*t
added RBFNet - needs still work
accidentally removed RealVectorRange
Changed Interface of KernelFunction a bit. Fully Implemented GaussianRBFKernel. Kernel and RBFNet are now working
updated the interface ob AbstractModel. Additionally derived shark::Exception from std:.Exception for Boost-Test. And the Feature Interface is now implemented using define SHARK_FEATURE_INTERFACE. Testcases are all running, but i can't check the untested code...
implemented LinearNorm-Model and began to implement ValidationError/CrossValidationError
fixed ML after deletion of Core/SearchSpaces.h
fixed ROC crashes.
tested LinearNorm, Partition and SoftMax. All Working. Nearly finished with Partition, only 1 Create... Missing.
reworked LinAlg a bit. implemented PCA
various changes
implemented LDA and LinearClassifier
found a way to workaround vector<bool> so that binary classification now uses boolean labels
added regularization to LDA and implemented multiclass FisherLDA
fixes
changed parameterVector to setParameterVector and added a header for linux-processtime
made a change to AbstractError such that now instead of calling risk() the old way is used, but the normalisation factor can instead be queried so that also the derivative can be scaled correctly
implemented classifier methods. should work, but the tests are not exhaustive...
new Version of Dataset. does NOT compile at the moment
further changes to the Dataset class. Design should be nearly stable now. Next is getting it to compile
reverted partial changes. Got Dataset to compile. everything else is currently broken
again i hate svn for net accepting my reverts...
got nearly everything to work with the new dataset. added regularization to linear reression. perceptron and svm needs fix
tested and fixed: LDA, FisherLDA and LinearRegression
Added configure to nearly all models. Changed the KernelModels to use the Kernel as a template parameter. Added IConfigurable to AbstractKernelFunction
fixed compile errors in dataset
enabled Perceptron again. all tests are now working
added serialisation to most models
finished ConcatenatedModel. It's tested and working
moved missing functionality out of partition. removed partition. testing not possible because of bugs in Dataset
added missing file
new version of Dataset. there is a very strange bug with the destructor of std::map. Try Tests/Perceptron.cpp to see it
kind of fixed the bug. don't know how, but it is gone.
Tested Dataset. Everything seems to work
tested creation of CVPartitions. working
tested cverror. seems to work
aded ROC
fixed some names...
missing file...
fixed fisher LDA again. eigenvectors need to be transposed, because they are stored in the columns of the matrix. Please don't buggify it again...
added resetting of parameters before the call to train
fixed creating subsets from subsets. added pack() which packs the elements of the (data)set
implemented derivatives for kernelfunctions and tested everything
SigmoidFit and JaakkolaHeuristic
added BalancedErrorFunction for balanced multiclass classification
added MatrixType to find the proper matrix for a given Vector
corrected types in AbstractModel and moved the standard derivative to a new subclass DerivativeModel. Also added detailed documentation
fixed RProp in Release mode (damn Compiler), added derivatives to Rosenbrock, fixed Rosenbrock
rough sketches for CrossEntropy
changed all standard derivatives to fit the new convention. working
first version of the new RNNet. does compile but will propably not work
fixed parts of RNNet, but still not working
further debugging of RNNet. eval() works now and the most basic setStructure(in,hidden,out) is also tested
added distanceSqr function as a speedup for sparse vectors
fixed a bug in distanceSqr
cleaned up FFNet interface and fit it to the RNNet interface
RNNet derivative is now working :). but needs still work
optimized calculation of the derivative, changed inputs to RealMatrix
final commit fpr RNNet :)
added boilerplate code to RNNet, changed FFNet partially back to reflect the changes to FFNet
searched and destroyed a Bug in RNNet
added Online version of the RNNet. Needs debugging, but it allready compiles :)
added a common base for the structure of both RNNs
OnlineRNNet is now working
changed type of the sequences used by RNNet to the new typedef Sequence in LinAlg/Base.h
externalized structure of the (Online)RNNet
removed TimeSeries
removed old EALib
removed old erroneous branch
removed unused src files
updated structure of tests
forgot to kill Tests/ML/
small bug in CG fixed. there is still something wrong
improved tests for CG and BFGS a bit
commit 64d437123b59c4e918241e1927bfdc49f62f7938
commit f2e030ee1366c3dbffcf47f6222caeb8c7587614
commit f4ec8c529514349321e089aef30515014c881793
templatized VectorStatistics and used meanvar to implement NormalizeComponentsUnitVariance
fixed VectorStatistics.h and added some tests. more to come
removed acciedentlz checked in eclipse file...
added Lasso. not working.
fixed linear model
cleaned up code of ObjectiveFunctions. Working
forgot tests...
fixed single objetive benchmarks and added evaluation counting to error Functions
updated a lot of Benchmark Functions.
removed and saved components which will nut be ready until release
also removed the tests..
missed a file agan :(
added a flag system for the optimizers and added standard header to all files in include/{Core,Algorithms,Data} and src/
fixed a bug in the dataset from set constructor and added new functionality: equality tests, creation of subsets from datasets or sets
added a SigmoidSwitch in FFNet which made all other variants obsolete
added the same sigmoid change to the recurrent networks
fixed a Bug in Dataset
added first part of documentation. fixed bug in Dataset
fixed some more bugs
next step of tutorial done. fixed more bugs. removed warnings
added stopping criteria (needs checking) and fixed some bugs. LogisticLoss will be removed soon in favor of CrossEntropy
improved error functions a bit
replaced shared_ptr with scoped_ptr
some fixes
fixed a compileerror in EarlyStopping (which will be removed really soon)
fixed serious bug in IRpropPlus
fixed a few compile error and warnings. statistics now have an iterator interface again
added masking mechanism to OneNormRegularization
made interfaces more stl compliant. added an example program for ELMs
added tutorial for the ELM
added function for solving linear equations
added NegativeLogLikelyhood
crossEntropy finished, therefore removed logisticLoss. added KernelTargetAlignment
added a not working test for CrossEntropy. There may be a bug lurking
added PCA example. fixed bug in PCA.
added Test for SquaredLoss
fixed Test for Loss Functions
cleaned up examples and docs. Hopefully i didn't select something needed...
moved StoppingCriteria
added Tutorial for CV Folds
added Tutorial for CV Folds(2)
added tutorials for uBLAS and Subsets
improved balanced cv-sets so that they now support multiple classes
somehow i forgot to check in some of my changes
fixed bug in RBFNet.h and improved documentation a bit
documentation
improved automatic kernel derivative tests
added sparse support to kernel tests
documentation
added data importing tutorial
added the RBM Library to Shark
added RBM Problems
fixed some include paths in RBM package...
changed problems a bit and added examplecode + tutorial
fixed numeric instbaility and linker error in detsymm
updated a test for logdetsymm
added derivative estimator for objective functions
added documentation, rewrote CMACMap, removed CMACFunction
added IRLS, our first second order algorithm. it's an iterative newton step solver
big sorry, i broke Rosenbrock...
fixed construction from two unlabeledSets for now
updated RBM and added parallelTempering
fixed bug in parallelTempering
improved numeric stability of the RBM. no more runtime warning
added Dataset fix
fixed serious bugs regarding subset from subsets in Labeled/UnlabeledData
removed some superfluos includes
added TruncatedExponential and fixed linker error
added Typenames in Factory
updated RBMs and put some useful functions into Math.h
added various fixes regarding shark tests
unified ErrorFunction and LossBasedErrorFunction
added elementwise functions for vectors, like exp or log
added testcases for serialization of models
various fixes and corrections
renamed evalDerivtive of the Kernel
adding proof of concept for generic uBLAS templates
fixed VS2010 bug regarding serialization
fixed PCA example to compile on VS2010
rework of FFNet and added tutorial
missed changes to LinAlg
forgot CMakeLists.txt
fixed old bug in FFNet regarding configuration
fixed a bug which went through my review...sorry
there was a test missing
i really forgot the tutorial..
added initializer framework, also put all functions in LinAlg/.h in the function space
removed old source files of Rng which are not needed anymore
updated documentation of ConcatenatedModel
fixed a lot of doxygen warning and used the initialization framework to make some implementations of parameterVector/setParameterVector easier
fixed bug on visual studio
fixed RNG testcases
incorporated tutorial changes suggested by matthias (thanks again)
various fixes, also new tutorials
fixed rng (hopefully)
added tutorial for general optimization and ublas
tried to fix 32bit test fails
this commit is only for tesfinding the bug in our continuous integration system. unfrotunately i don't have the means to reproduce the bug at home. will remove the changes later
reverted previous changes and added new change. test will fail expected
commented test out. maybe the rest works
commented more parts out
commented some parts back in
commented some parts back in
commented one more in. i think this will fail...
commented one more in.
bug, be gone
added union of flags
added some std::
fixed dataset subset tutorial
removed old rbm module, new is coming shortly
added new RBM Module
some minor changes
forgot a file. sorry for breaking the build
another remote debugging session
hopefully, this fixes the bug
some debug outputs. i hate remote debugging without debug information
some debug outputs. i hate remote debugging without debug information
some debug outputs. i hate remote debugging without debug information
fixed the first tutorial
fixed another example
updated another tutorial
updated another tutorial
fixed rbm documentation and tutorials
added RBM files to makefile
forgot image file
added brief documentations
removed some debug output
somehow fixed the test
some fixes and documentations
reduced a error level of a test
added test for marginalization
tested conditional probability on truncated exponential
fix
fix
fix
fixed 1 bug and 1 senseless warning
added missing subset method
fixed beta for gaussian rbm
added a throw at usefull position :)
made the code a bit shorter
added another throw at useful position :)
fixed overflow bug while choosing between layers with many neurons in partitionFunction
forgot a debug line
fixed small typo
hopefully this fix works
please let it work now...
fixed compile errors
added Parallel Tempering
added a few definitions and made the test more usable
added defines for gaussian bianry rbm
added defines for the truncated exponential RBM
hopefully, this fixes the spurious bug...
fixed problems with the last checkin. also made the Energy function cleaner to use
changed LDA to use cholesky decomposition
added small changes to the FFNet and other small fixes. also renamed the negativeLogLikelyhood with the precomputed partition function to negativeLogLikelihoodFromPartition because of compiler problems
fixed a bug in the RBM module
accidently broke FFNet
added -fPIC to static build
added fast_prod, needs benchmarks though. also fixed 2 test cases and reactivated all tests
missing unit test fixed
sorry, forgot to check non atlas code...
tried to fix compile error with windows
disabled useless warnings in MSVC
again tried to fix isfinite problems
maybe this test now works again?
replaced a few more isfinites
replaced a few more isfinites
made invertSymmPositiveDefinite faster using ATLAS
made atlas a bit more stable
added first blueprint for automated project generation
small fix
fixed FFNet...hopefully
fixed
fixed
made fast_prod faster and tuned LDA using ATLAS
added ATLAS support und optimized Linear Regression
fast fix for jenkins...sorry, will change LinAlg later
accidentally checked in test verison...SOOO sorry
incorporated some fixes. hopefully it runs now on windows
LDA + fast_prod
used fast_prod to speed up rbms
there were some typedefs wrong after the recent update of CD
updated pca and ffnet
may bad
debug output for QuickProp
removed debug output for QuickProp, fixed test
fixed test
splitted LinAlg in separate files
made the CMAC Test faster...
added documentation for the new LinAlg operations
improved GaussianProcessEvidence & LinAlg
added some std::
added some std::
added some std:: and cleand up WeightedSumKernel
added some std:: and cleand up WeightedSumKernel
one more std::
wrong order of arguments...bad bug
removed some warnings and added some more std::
removed one more warning
fixed a warning in RBM module
trying to get the test to work
kernel tests not working. old test wa sbuggy
fixed small bug in gemv
rewrote kernels to be stateless and offer intermediate values
fixed two failing tests...sorry for that
hopefully fixed
some changes to the RBM
creating folder for the new batch interface
adding the subfolder
creating branch for the batch interface so that current users won't check out a bad version
new Dataset. won't compile. nothing works.
small issue fixed
fixed a gazillion compile errors
more fixes (nearly compiles now)
added docu
fixed some more errors
fixed parts of Data/UnlabeledData to actually work now :)
got LDA up and running
fixes
ported changes to the branch
improved fast_prod for vectors (uBLAS is scary...)
added changes from trunc
fixed a dumb bug...
ported to branch
nearly final answer on fast_prod for matrices implemented. it is finally even readable
small fix
missing reference
small fix
still something wrong. sorry, these issues only pop up on the other branch...
haha i will never optimize things, everything broke :)
hopefully fixes windows errors
small fix
first draft for part of new RBM
various changes to the batch brnach
fixed fast_prod and optimized it when no atlas is activated. also added benchmarks to the test so that you can check, whether atlas is used properly
last fix to fastprod and emrged changes to the branch. also tested CrossEntropy partially
merged last fastprod fix to the trunc
tested squared loss
cmac update
ported ConcatenatedModel
some more ports and fixes
fixed meanvar
fixed VectorStatistics
converters
ported data importers
ported ncll test
commented out the slow benchmarks in debug mode
merged stuff, added BlockProducts(wip)
further merges
various changes
FisherLDA running
SoftNearestNeighbor running
NoisyErrorFunction working
CVDatasetTools half working
CVDatasetTools working
ported GaussianKernel and AbstractKernel. also ported absolute loss
partially merged changes from trunc until r1966
merged other changes and fixed some small bugs + improved blockedOperations
sucessfully made the gaussian rbf kernel fast
merged changes from trunc and further improved the gaussian kernel
replaced std::fill with std::copy
some further rbm changes
completely broke RBMs :). but it is nearly ported and must only be fixed now
broke a bit more RBM code :-)
small changes
rewrote LinAlg Traits
rewrote LinAlg Traits further. now can also traits for compressed matrices be added
convenience interface for batches and a bit improvements for the rbm
added missing files
fixed a heapload of compile errors for the RBM. also added a new function to efficiently sum rows or columns of a matrix
small generalization for different value types
down to 56kb error messages in RBM code
fixed further errors in TemperedMarkovChain
RBM now has < 40KB error messages ;)
oops
RBM compiles now, but doesn work properly *sigh*
got a few tests to work, still not working, though...
RBM_Energy works and is substantly improved
added test for the partition. working
Fixed Bug in log likelihood
merged some parts from the other branch
there was an update while merging...
added more tests for VectorStatistics. There was an actual ug found during this...
removed some parts of old RBM code
fixed a bug in AverageGradient. now PCD CD as well as PCD are working properly
PT is working
changed size of PTTraining back to a smaller size which is faster to test
further speed improvements for the RBM
oops wrong paramete rin the test - would have been damned slow in debug :)
gaussian layer working :)
ported most of the GibbsSampling test
two more RBM-Tests working. also started porting the TruncExp neurons
RBM is now completely ported and stable
small changes to make the code more readable
added indexed iterators as well as a few smaller interface changes to the dataset
changed batch itnerface to base on boost::range
added missing files. also added functional.h which is intended to be a place for basic algorithms used throughout shark
investigating performance issues in debug mode
there is apparently no bug in get(batch,i). still slow in debug...
eliminated a possible error
tried to fix another issue
maybe this fixes another errror...
changed suRows a bit
made suMRows always working :)
made sumRows always working :)(now really). And added a small test for summing a complete matrix
commented out all examples
hopefully fixed the compile errors and a boost bug...
missed a file
fixed Energies of the gaussian neuron layer
merged changes from Trunc into branch
started Rework of RBFNet and implemented a matlab like repmat function repeat which repeats rows in a matrix
RBFNet::eval now works, weightedParameterDerivative compiles, but fails the test
RBFNet working. gamma are now log encoded
missed to remove an unneeded variable
KernelNearestNeighbourClassifier working
KernelNearestNeighborRegression working
KernelExpansion as well as the GaussianKernel work and pass the test
changed propability to compute the log probability
updated calculation to use long double
merged changes from trunk
whoops
changed definition order of distance to be after distanceSqr
another swap
fixed another syntax error
fixed some type-magic-errors
worked a bit on the binary trees
go NearestNeighborRegression to work with a full rewirte of the BinaryTree/KDTree. other trees are still missing
small beauty fixes
LCTree seems to work...
all Trees working as well as KernelNearestNeighborClassifier
SoftNearestNeighbor working...so trees are done
cleaned up Core by moving some stuff into the new Core/utility/
got most examples to work
removed Counter
three more examples are working :)
fixed Linear kernel
Monomial Kernel working
tried to fix an issue with sparseness and transformations
fixed comments a bit
PolynomialKernel works also got a step further in my crusade against uBLAS
Polynomial kernel also works for sparse...awesome
various bug fixes in no-atlas mode. disables warnings to get LDA working. added QR decomposition. fixed various bug in LinAlg along the way. added a lot of unit tests
missed a file
restructured LinAlg to be less of a mess
missed a file
some fixes for the truncated exponential and fixed the buildscript
speeded up tests a bit
further speed adjustments
draft for Errorfunction for FFNets with sparse hidden units added. does not work.
forgot a file
fixed PCD test
sprase ffnet error compiles but test is still only stub and auto-fails
implemented faster moore penrose inverse. couldn't test it because i forget to checkin some of my changes on the other pc...
forgot to implement permutation
added missing files
rewrote the triangular solvers and circumvented some uBLAS related bugs
g_inverse commited and workin. but i think there is still some weird bug in pivoted cholesky...
LDA can now solve singular problems
fixed small bug in LDA
small first bug fixes for Sparse FFNet
fixed obvious bug in the Dataset
fixed SparseFFNetError. should work now
yeah of coruse, you should always check whether the returned error is correct...
damn you, numeric.
...
more numerics
fixed a small bug which was bad with very big datasets
merged changes from trunc
added backend for gotoblas. made everything awesome :)
sorry
fixed compilation errors
improved support for OpenMP and added parallelizable KNN-Classifier which does a brute-force search
fixed some compile errors. sorry
added more tests and fixed a minor bug
small change to be sure...
testing version
another testcase
even simpler version
working
oops
oops
tried to fix gotoblas compilation
added gotoblas support to the automatic generated example CmakeLists
slight error
added DenoisedAutoEncoderError. It's awesome
missing File...
added exact gradient calculation for small problems and added a few tests for different components
fixed another bug in ExactGradient
missing file
merged changes
fixed some warnings
ported Scaledkernel
fixed small bug in second derivative of cross entropy. added a new test
fixed a weird bug
fixed a debug bug...
fixed compilation issues with gcc 4.7.0 also fixed bugid 3554077
added missing file
fixed some warnings, a compilation error with gcc and tested some examples
various changes. WeightedSumKernel does not work, though. alsoa dded first steps for the right MKLKernel
forgot to check in a fix for a test...
added a new ProxyReference for vectors which allow conversion of all Dense-Storage-Vectors to this Proxy. This solves the copying issue of subranges or matrix-rows
committed RBM changes and fixed the GaussianKernel test
ported a few mire kernels, changed the kernel design and aported most parmost of the svm stuff
ported a bit more
commit ef5d7a4ebbe8e47b5e7ac74cd55e83ea71b0c03e
merged changes from trunk. apparently the changed LibSVM Test is not working for me...
sorry
sorry again
stray intermediate
improved DataView and cleaned up a bit
hunted a lot of warnings and ported CVError
hopefully fixed doxylink errors. Also remove superfluous template parameters in AbstractTrainer
changed dataset a bit...
sorry, debug line prevented test from finishing
fixed some meta-programming bugs and ported RF-Tree also improved DataView
tried to add a bit of documentation to the batch interface
improved test
fixed RFTree
ported NBClassifier
ported NegativeGaussianProcessEvidence
adopted roc partially
first try on rewrite of KTA
fixd KTA::eval, still evalDerivative not working
KTA is now working
first go on OnlineRNNet...failed. still compiles, so in it goes
ported onlineRNNet
oops
ported RNNet. activated a lot of tst. currently 132/132 passes. takes 156ssec release with ATLAS
merged changes from the other branch and ported most of the code
ported MKL und MultiTask stuff. InputDerivative in WeightdSUmKernel is commented out because of issues discussed via mailing list
ported NBClassifier trainer
ported LOO-Errror and SigmoidFit
removed unneeded files
whoops
fixed derivative issue for weighted sum kernel
improved kernels, clened up LinAlg and implementd a subrange kernel
ported RadiusMarginQuotient
updated Dataset and redesigned NN-Search
added a bit documentation of models
added missing models
ported changes in other branch
added docs for kernels
whoops
obvious typo :/
added documentation of Cost/Loss and improved the interface at the same time
another obvious error
forgot a Loss function
added optimizer interface description
small improvement
removed obsolete early stopping
addd documentation on stopping criteria
addd documentation for Trainers
rewrote CV
got remaining examples to work
starting clean up. replaced noVariables by numberOfVariables in AbstractObjectiveFunction to conform with the rest of shark interfaces
moved objctive functions from Core to ObjectiveFunctions. clean up goes on
copy paste error
trying to fix the fusion issue
moved optimizer interfaces to Algorithms. removed TypedSolutions
moved ScopedHandl to utility, got hdf5 working
moved regularizer and changed include guards to reflect the new positions
whoops
fixed small error
further error
whoops
updated documentation
fixed a bug
fixed Csv Test
fixed an unitialised memory issue
Whitening now uses pivoting cholsky. also renamed LinAlg.h -> Inverse.h
small fixes
commitioremoved a few problems with VC10 and fixed a lot of warnings
small fix
some fixes to the docu
added CG solver for systems of linear equations
hopefully fixed bug on GCC4.21 and VS10
working on VC10
small change which fixes som issues on VC10
my bad
updated Metric.h
small fixes and improvemnt of documentation
changed tolrance to 1.e-13
removed TypeErasedSingleObjectiveOptimizer and changed optimizers in DirectSarch to adhere to the interface directly. also changed example and tests when necessary.
kernel oc improvement
kernel further doc improvement
further doc improvement
fixed somee smaller errors and wrote documentation regarding stopping criteria
reworked the first block of algorithm tutorials
improved further tutorials
changed interface for matrix blocking
updated dataset documentation a bit...
improvd dataset docu a bit more
updated tutorials on derivatives
updated normalization tutorial
updated Models to have a stateful interface
added very expensive test to chck that the CD gradient is corrct - i calculate the *exact* CD gradient
fixed error in RNNet
added flag for parallelizable objective functions
broken, need to fix it with gdb...
tried to fix errors in VC10
slight changes to get a real mdian function for arbitrary sets
hopfully fixed shuffle
hopfully fixed shuffle
updated ModifiedKernelMatrix
hopefully fixd prolem with big sparse datasets
changed KernelDerivative Interface
fixed NearestNeighbor
removed some interface errors and reduced testing time in debug mode
fixed som warnings
renamed Fermi to LogisticNeuron and other neurons have now 'Neuron' at the end
improved KMeans to use the batch-interface
applied the change to the algorithms
solved compilation problem with boost 1.46
moved the implementation of some optimization algorithms to cpp files
renamed split
updated parts of the documentation
moved batch branch to trunk
renamed split again to splitAtElement
updated data interface
fixd dumb bug.
restructured tutorials
fixd intentation
tried to remove todos
hopfully fixd pca bug
fixed PCA and addd a new test case
sucssfully updated the first part of the dataset tutorial
reordered CMakeLists.txt a bit and splittet into tree parts
removed dataset constructor, added free function instead
fixed clustering bug
fixed a small bug in splitAtElement and got serialization to work
updated documentation
further cleanup of data docs
updated elm tutorial
changed LibSVM Test a bitbefore new parser is integrated
updatd Libsvm parser and fixed a bug regarding dense vector proxies
fixed problem that sparse matrix rows are always copied when arguments for kernel::eval. huge speedup for sparse-SVMs
changed problematic output from examples to cout also other small fixes
fixd bug regarding swap of elements
updated shark packaging for cmake
fixed small bug in cmake file
tested basic kernel matrices
fixed compile error on some compiler
enabld multithreadd error functions...yay
fixed compile error and a few warning on MSVC
fixed small configuration bug in FisherLDA
fixed OpenMP on MSVC10
incorporated better box constraint handling mechanism and cleand up DirectSearch a bit
forgot a file
applied small optimisation to the kernel matrix such that we now use the direct iterators to the elements instead of the dataset iterators. good improvement in performance as a result
fixed Bug in Matrix-Proxy and Benchmark test
fixed compile error in Lasso
allowed parsing class labels 1.0000 for LibSVM import
slight changes to LRUCache
wrote tutorial about writing objective functions, took the chance to chang the interface again. should be working. also update objective function tutorial
moved slow to compile libsvm importer in library
included LBFGS Algorithm
included LBFGS Algorithm
fixed broken build
merged the new binary svm solver
fixed ugly bug in KMeans
fixed warning in new Svm solver
tested CachedMatrix and found a small bug :(
partially fixed performance issue in LinearModel with sparse arguments. it is now working for the McSvm-Case (dense output, sparse input)
imported ublas to shark and removed most of it's features
sorry :(
altered the svm solver another time
hacked in support for different regularization parameters
changed order of regularization parameters...
reworked diagonal matrix
changed shrinking another time
fixed an error and warning in debug mode
fixed two more warnings
fixed a LOT of warnings in ublas
fixed last ublas warnings
removed another 3400 lines of ublas code
another intermediate step
fixed small bug in Dataset
fixed a small bug in the libsvm import and added a bit of parallelization
various cleanup
fixed compile error
checked for and removed warnings on MSVC 10
rewrote parts of the RBM package
added .gitignore for git users
improved computation time for dense vectors
added new dense kernel optimization
added new functionality to C-Svms
small speed improvement of jaakkolla heuristic in th sparse hack. it's hacky though
Added Sørens algorithmic improvements
hopefully fixed an error in C++11
removed temporary files
add possibility to modify the linear part of an svm solution
sorry for compile error...cleaned up my mess. also fixed small error in nw example
fixed error in example
fixed an error introduced by the change of the line search behavior
fixed Loo error. sorry
added special case for rank deficieent SVM problems
fixed shuffle on VC08
added squared hinge loss C-SVM and an alternative way to compute the bias
fixed a small interface error in the distributions
added automatic code snippet inclusion to documentation
some fuckups with git...
ported the first_steps tutorials to the new system
ported tutorial code as well
bla
fixed stupid bug in CMA
fixed stupid bug in CMA
ported some more tutorials
fixed out of source builds for doc
started to rewrite linalg tutorial and fixed lda
various improvements to the linAlg tutorial
renamed tutorials .cpp->.tpp and automatically strip the tags from them with a conversion tpp->cpp#
fixed a problem with broken ;
fixd problems with regenerated examples after cmake run
updated .gitignore
hopefully fixed the problem with shuffle
adapted filenames to point to the new examples
fixed tut2rst such that cmake now fails when the run was not successful
updated elm tutorial
updated ffnet tutorial
fixed broken cma tut
ported and updated CART tutorial
updated pca tutorial
fixd bugs in MultiTaskKernel
ported kmeans and rf tutorial
ported svm tutorial
ported CSVMGridSearchTutorial
ported Sparse Autoencoder tutorial
just renamed the rbm module as it does not contain any code
small cosmetic changes
fixed case for svmProblems that for the subproblem of the workingset the second derivative was numerically 0
added new McSVM solver for box constrained svms (split from the big one)
added temporary fix for boxconstrained SVM problems
various fixes
fixed various bugs in nearly unused shark code
changed naming scheme of variables in the trainer to match the standard m_ scheme
unified shrinking counter in QpSolver.h and fixed a hard to find bug on the way...
solver now returns the unpermuted solution and uss the QpSolver interface
random clean up and ported the LLW trainer to the new solver
cleanup
fixed interface error with Linar Mc Svm
fixed very small glitch when handling deactivated variables
fixed bug in radius margin quotient
added some new cool operators to vectors
added new helper-function to partition a given number of points into batches
small cosmetic change to kernel expansions internals
rewrote csv import
removed a warning in lassor egression - the timing code was unused
oops forgot to fix the tutorials...
added special case handling for gaussian kernels and sparse data
fixed strange error. sorry
fixed small error in const ref handling
used same label handling as for libsvm
added kernel kMeans
repwrote LinearClassifier to be an argmax converter of a linear model. works great
fixed install of examples
typo
added stStructure interface to KernelExpansion
removed default choice for wSS slection in CSVMTrainer
cleaned up LinALg and removed unused algorithms. also fixed a bug in CSvmDerivative computation. Also moved everthing in LinALg to namespace blas
fixed documentation warnings throughout sharks
oops
removed One-Hot encoding and fixed bug in BFGS
fixed a few smaller bug
fixed compile error temporarily
removed thee automatic experiment interface for MOO experiments.
removed Logger
cleared Core from all now unused concepts
added a framework for LineSearch Algorithms. Reimplemented CG using a defined formula (with reference)
reduced memory usage of RFTrainer
added parallelization to RFTrainer
added exampl code for brute force KNN CV
fixed small bug in the LinarSVM
Linear svms now use linear classifiers. also removed necessity to initialize model prior to training
missing test
added KernelClassifier and let all classifiers use it
fixed bug in Dataset Iterator
improved KrnelSGDTrainer a bit
added new SimplexSolver for McSVMs which fixes a lot of the issues of the old solver. working
aded a solver for systems without full rank, hopefully fixed LinearRegression by that
fixed CSV bug and added tests for it
added more tests, thus fixing ticket #59
fixed compil error and test
removed KalmanFilter
cleaned up LinAlg further and moved lineseach algorithms to Algorithms/Impl. Linmin and Lnsrch removed as they are not needed
removed g_inverse and thus th last remaining invert* function. the several version of solve... are to be used instead
removed now unused ATLAS bindings
small interface change
fixed compile error for export_csv
added fucntion to compute the complete kernel matrix at once
added default batch sizee argument to libsvm import
updated example buildfile such, that changed tpp files are automatically converted to cpp
fixed tpp2cpp target dependencies
tried to fixa few warnings and updated KernelHelpers.h
made csv import easier to use - when a whitespace is given as a separator this counts as no separator, so there is no need for a 0 as argument anymore. Also fixed a bit of typos...
improved LibSVM import to use less memory
added measurement of wallclocktime to timer, which is now the default
new branch for linear algebra rework
initial commit of the new linear algebra library
small fixes. Thanks Tobias
cleaned up gemm
optimized QpMcLinear.h and tested performance of LinAlg in pure sparse case
ported LassoRegression
improved LinearModels derivate
fixed an error whn swapping rows between sparse matrices. shuffle is now working
removed zero() and replaced all occurences with clear()
moved Tools/ to LinALg. renamed ensureSize->ensure_size and incorporated in ublas
moved most of BLAS/ to LinAlg or BLAS/ublas/
finally moved BLAS/ublas -> BLAS
rewrote LinAlg completely
fixed comment handling in csv.h
added operator *= to dense vectors and matrices allowing for elementwise multiplication. replaced element_prod in shark by * or *=
fixed compile bug for older boost versions
optimized Gaussian RbfKernel derivatives
more missing includes
more missing includes
added appending of datasets
fixed an issue regarding serialization of Classifiers
small cosmtic changes
added second derivates for losses
added proper prior to LDA
added a bit of numerical accuracy to scaling of C
added an ATLAS binding for inner_prod
improved handling of nested proxies
fixed small bug in sparse matrix transformations. deactivatd failing tests and declared as won't fix. Fiing that would either be stupid or slow - even in cases where it wouldn't require a change
fixed two errors on ubuntu: a missing include and a numercial problem with PCA with verz small eigenvalues - they are now ignored as they are only numerical noise
fixed small bug when adapting C while using cross validation(no current result is invalidated). Also removed a specialisation that is not necessary
fixed a bug in jaakkola heuristic which occured when two identical points had different labels. Also rewrote it so be a lot faster when used with ATLAS. still needs some better optimization
fixed serious bug in LibSVM WSS. Does not effect final results, and does not effect gaussian kernels. Only runtime is affected. Wrong matrix diagonal was chosen.
made CrossEntropy numerically more stable. Also changed the starting step with of KernelSGDTrainer to be more reasonable
fixed dysfunctional nearest neighbors test.
further fix which fixes the learning rate for high values of C
added function for numerically stable calculation of soft_max(x) = log(sum(exp(x)), where x is some vector
added AIS and other algorithms to estimate the partition function
made CD and estimation of partition function multithreaded
fixed small bug introduced by too small type
restructured code a bit to allow estimation of several algorithms from the same sample
fixed small problem that too many threads are created when there are not enough batches
added burnin time
added the full propper AIS as proposed by NEAL
added multi threading again
fixed broken example
changed softPlus/soft_plus to use floating point precision. had to tune tests to the new numerics
fixed temeprature handling and renamed algorithms
fixed legal thingy that prevented CPack from building
removed GLPK.cpp as it was forgotten..
added mean classifier
rewrote RFClassifier to use MeanModel
hopefully fixed vector for bool arguments
added a stub for a weighted error function
partially added code from Kang
fixed a bug with reading in files with single integral values
added feature selection to dataset
bit of clean up
various clean up and added flip-thestat sampling
various fixes and simplifications throughout the RBM library
added HingeLoss Loss-function
added squared hinge loss
cleaned up MarkovChain Tests
implemented vector valued min max to single scalar values as in max(vector,0.0)
sorry for compile errors
tried to fix normalized kernel test (was unused before). removed now unneeded old kernel tests, they are all in new files by now
reactivated two older tests and got them to work again
removed FitnessTraits
fixed small bug in LDA
added psilonHingeLoss for Regression
added a bit of doc to the losses
partly rewrote DirectSarch
sorry for the compile errors in the example code
added missing include file
got DirectSearch back to compiling and cleaned up a lot of things
updated CMA a little
I broke PT and fixed it again. Also added a way to store the energy differences for running chains using a wrapper for PT
rewrote ElitistCMA
forgot to update file header
removed two unused files
oops
removed more unused files
added iterator constructor to vectors
further clean-up of direct search. moved a few deprecated components in Save before they get deleted. Merged all elitist cma parts into the CMAIndividual and rewrote MOCMA algorithms to use this
missed a file
further rewrote DirectSearch
unified MOCMA and ElitistCMA code
made MultiVariateNormalDistribution simpler
further cleaned up MultiVariateNormal and added MultiVariateNormalDistributionCholesky which uses a cholesky decomposition
fixed dataset tutorial
added cholsky update
added author/date
added parallel mini batches to the RBM package
removed AbstractFactory as well as unit tests for it
removed plotting function
removed Derivative.h
removed Statistics.h
removed docs to removed classes
updated ignored files in the doc
removed SharkDefs - not needed anymore
moved a test file tool to the Test folder
cleaned doxygen documentation a bit by removing template specialisations and deciding that detail/* and Impl/* should not be documented. Also moved stuff around to use this
redo of commit that removed Statistics.h
removed the Batch bindigs for boost::fusion::vector, they are superseeded by the other methods and thus useless
removed SearchSpace and IsVector to make things simpler
fixed a broken test and cleaned up a small thing which allowed me to remove CopyConst. Also figured that SharedVector is some remains of some old thing we don't ned anymore -> can die
moved Fuzzy to addons
added missing documentation of statistics
oops
error in gitignore lead to ignoring the examples
added the first part of the SVM approximation
merged changes
added test for the box constraint handler
small improvements to KernelBasisDistance - added a method to return the computed beta value
cosmetic fix
added a Test for PenalizingEvaluator and fixed a bug
added Dominance relation Tests and small fixes
fixed two reported bugs, one crippling performance, the other crippling the MOCMA. added unit tests and fixed two other smaller bugs. MOCMA has now a unit test which passes
checked ZDT implementations and benchmarked the MOCMA with two more functions
checked and cleand up benchmarks again and added even more tests - now also for SteadyStateMOCMA
checking the other functions as well and cleaned up the code
added unit test for RealCodedNSGAII and fixed small compilation bugs. ZDT2 fails however
whoopsie
small cosmetic fixes and made tests a bit faster
fixed a bug in automatic return type promotion when the argument was a scalar in binary operations. now vector<int>*double->vector<double>
added example for experimental usage of MOCMA
added an example for an MOO Experiment
unified mu_plus_lambda and mu_komma_lambda selection, it now fits in the general interface
clean up, improved a test and removed a dysfunctional one, which only pretended to be a test
replaced RBFNet by RBFLayer for better reusability
added input derivative for linear model. this is needed to create a RBFNet using RBFLayeer and LinearModel
added possibility to only train one part of the concatenated model
added a model representing a convex combination based on the lienar model. the main differences are the constraints used to enforce the convexity property
added first draft of negative log likeelihood function. This is not stable as we explicitely compute first p(x) and then log(p(x)). as p(x)->0, this becomes unstable
deactivated for now the fast covariance update as it is numerically unstable
added numeric safeguard to covariance matrix update, checked that it now runs finee with 200 dimensions
added efficient rank 1 update for cholesky factors
added new, fast, intelligent cholesky update which does not require the inverse
a test was failing after the ellipsoid function got changed
changed the Evaluator interface a bit so that it is possible to evaluate groups of individuals. This might allow us later to introduce noise handling
fixed small fuckup during another fix... thanks francois-michel
fixd problems with dynamic linking on linux
added an evaluation example for the CMA
made an example on how to train a mixture of gaussian unsupervised using KMeans and MaxLogLikelihood. Works on the faithful dataset, so i am sure it will always work...
oops
oops git failed me
removed bad implementation of Noisy autoencoder error
made it possible to set the initial step size
added distribution which samples image patches from images
added a testing function whcih checks that the weightedDerivatives method produces similar results as the individual calls
optimization for WeightedDerivatives not to call derivatives neadlessly
added helper function to interpret vectors as matrices more easily
added model which corrupts its input by gaussian noise. usefull for denoising autoencoders
improved temporary proxy handling
rewrote FFNet and fixed a bug that lead to a crash when the bias was not used
1. fixed small bug in that the maxIteration count was notused in one version of CG. 2. replaced Gram-Schmidt orthonormalisation for creation of random rotation matrices with househodler reflection algorithm, which is numerically more stable - this is now good enough for benchmarking numerical routines in high dimensions - it is also faster.
added a unit test for eigensymm which is at least a little bit more sophisticated. Also made a bit of the test code slightly faster and changed an unobvious noterror to an obvious correct piece of code
1. changed dynamic/static once more. now it should actually be working. 2. added bindings to ATLAS-LAPACK if ATLAS is build with full lapack (new build option for now, until i find a good way to query this)
moved the rank one update into the MultiVariateNormalDistributionCholesky
added new weighted datasets
added the possibility to query the batch sizes for repartition
small fixes to the Weighted Dataset
rewrote problem constructor to support for individual weights and C values for each class
added interface for new weighted trainers
added support for weighted CSVM training
added bootstrapping
added cod for the multinomial distribution
added another test
small fixes and added querying functions as for normal datasets
added weighting for LDA. Only problem is, that i can't get the unbiased estimates right
uncommented a test case..oops
removed DataObjectiveFunction
fixed small possible compile error
updated documentation a bit
added function which automatically rotates another (probably axis aligned) function
added triangular matrix-vector product: triangle_prod(A,x) which computes x=Ax
fixed a bug that Aydins build server found. I promise, i will never change a line in that ugly code
added faster implementation for trsv in certain cases
fixed HDF compilation warning
added loss for sequences and removed the derivative interface for AbstractLoss
fixed initial step size of CMA in high dimensions
added two Large scale CMA variants. need documentation/tests but working
fixed learning rate
definitions of cigar and discus where switched
reenabled hand-modification of eigenvalues
fixed bug in active covariance update
changed definition of beta such that now beta=0 is feasible for gaussian layers. accordingly it is now possible to hav a canonical definition of AIS etc which use equal temperature spacing. These methods are added for easier usage
some changes to add momentum and make some hyper parameters more reasonable
added broader support for regularizer
added the huber loss
added Tukey's Biweight loss for robust regression
added tutorial to train a deep neural network on a simple toy task. Added various helper functions to make this easier
added tutorial for deep ffnets and changed a few other tutorials by removing lying parts/updating them
fixed
added min function...
added regularisation methods to NoisyErrorFunction
added TiedAutoencoder (autoencoder with tied weights)
fixed LinAlg headers
added ZCA normalisation
added Dropout
added 'Impulse' noise for denoising autoencoders on binary data
added also simple autoencoder with the same interface as TiedAutoencoder (and sharing much code unfortunately...). Tsted and works
added Rprop with full weight backtracking
fixed svd problem
fixed small bug in tutorial when reading the datast. Thanks at Michiel
remove NegativeClassificationLogLiklihood and updated tutorials to reflect this. Had to add a quick fix for SigmoidFit
brought the Sparse-AE tutorial closer to the current state-of-the-art implementation
removed the choice of ouput neurons for autoencoders
-undid the bad decision to have no output neuron type in Autoencoders
added basic autoencoder tutorial
added denoising autoencoder tutorial
added RBM tutorial and fixed some compilation bugs. Sorry for the late commit, my local git repo was broken...
updated tutorial with probability of samples section. also improved the softmax model as to also work with only a single input
added mask to TwoNormRegularizer
added helper functions
added fix in case that the highest index is available and a separate index is givn
added matrix sin/cos for completeness
various changes
fixed missing function problem
fixed reported compile errors on VC2010
fixed the VD-CMA Algorithm. Worst problem was an error in the paper. I think it is now working fine, at least the convergence looks reasonable
fixed bug in tutorials page
whoopsie
fixed openmp issus with VC2013 as in ticket #80
fixed numerical error with NegativeAUC
added new batchwise cross validation folds
fixed small bug in tutorial
fixed compilation problems with boost 1.57 beta(which does still not compile because of serialization)
removed unusd components
Added the trigamma Function. Wow!
redid approximation of trigamma and added tetragamma function.
typo
fixed noalias(x)*=2 failure and added a matrix_set on the way for Data<RealMatrix>
added ModelKernel class which computes a kernel with a model to map the inputs into another space
changed name to reflect name of inner function
added optimisation for column_major matrices when an outer_prod is applied
various optimisations(and a small fix) for ElitistCMA
after i made another test more strict it turned out that this class didn'tset its proper flags...
fixed possible error with parallel Dropout
added new triangular matrix class with packed n*(n+1) storage, also cleaned up on the way
fixed warnings and activated a missing test
added push_back for vectors
added scalar add/subtract for vectors
missed a header file
added regression capabilities to KTA
added multi thread support for KTA(making the class even uglier)
added atanh
oops
fixed a number of warnings in release mode
fixed another warning
added functor for fitness sorting
small fixes in docu and added operator+= and -= for elementwise addition of a scalar. + and - where already there
added new benchmark functions
added alpha parameter to Cigar
fixed bug in CMA-ES. sigma was updated to early. Thanks Tobias.
put some band-aid around the itching serialization wound
shark compilees not with boost 1.57
cleanup
added pgm grid export for dataset and fixed compile error in SoftClusteringModel...noone has ever used that
small bug fix in debug mode that prevents class usage
oops
changed Softmax membership function to a more numerically stable default
fixed brain fart
undo change completely
added function which returns distance of points to all centroids
added scalar pow
oops
oops
removed IConfigurable, added AbstractMetric and cleaned up a little
added metric support to nearest neighbour algorithm
added push_back for batches in dataset
oops
removed superfluous template arguments from error function
removed superfluous template parameters in NoisyErrorFunction
added some direct encode/decode functions
some small fixes
clean-up and added macros for dll export. don't know about windows, but at least still compiles on linux
fixed missing include on windows
temporary(?) change
temporary(?) change
fix of makefile
temporary(?) change
various small dll fixes (and removal of warnings)
some bug fixes on 1.58 and also ensured that objective functions initialize their evaluation counter to 0
hopefully fixed MklKernel
hopefully fixed MklKernel...try 2
interface changes to AbstractObjectiveFunction::proposeStartingPoint and AbstractOptimizer::init. Also updated docs and did a bit of a clean-up while on it
Added TrustRegionNewton and removed the IRLS, which was a simple newton step. Also cleaned up a little bit more
added prerequirements for convolutional rbms
changes after code-review
added TwoPoint StepSize Adaptation rule
added convolutional RBMs. seems to be working
fixed some tests
updates step size adaptation algorithms
testing of new makefiles
updated latest changes of RBM code
fixed lapack issue and deactivated ScopedHandleTest
added new example project
fix svn ignore
fixed stuff
oops
changes to documentation. also changed atla to be on by default
fixed a few smaller isues with NDEBUG
fixed policy error
small fix
further updated documentation
a bit of clean up on the simulators
removed documentation warnings
fixed a weird warning using casts
removed empty directories
renamed Version to SharkVersion as this is a better name for linux...
renamed Version to SharkVersion as this is a better name for linux...
still forgot one occurence...
testing different cmake-file location
added info about arch linux packages
simplification of deb generator
Debian naming convention enforced (i hate cmake)
rudolc83 (2):
commit aa6a55c25a2443692c0e01a07d8f117022654bd0
namespace of isnan fixed
t_voss (327):
Copied Shark to trunk/Shark.
Incorporated feedback for second release candidate of Shark 2.3.0.
Deleted unnecessary files and incorporated feedback for second release candidate.
Accidently missed the no longer needed Makefile for SharkConfig.
Reworked ini-logo and evolutionary loop.
Last changes for homepage.
Ported back changes from trunk. Made Find(OpenMP) optional. Added testing facilities to the main CMakeLists.txt
Moved actual library to trunk.
Created branch SharkReworkBoostRng.
[Shark Boost Branch] Finally merged in changes.
Reverted changes to CMakeFiles.
[Shark Core] Initial check in of core components.
[LinAlg] Added typedefs for matrix proxies.
[AbstractOptimizer] Initial checkin.
[Shark trunk] Preparations and corrections for inclusion in debian.
FIXED: Compile-bug in CrossEntropy.
[Shark] REMOVED: XML parser with incompatible license.
[Shark rework] Statistics, initial checkin.
[Shark] Fixed copyright issues.
[Shark] FIXED: Bug in RMMEDA unit test.
[QuadraticProgram] Removed reference to std::tr1::unordered_set.
[Shark 3.0] Added EA component.
[Shark 3.0] Added subfolder for search spaces. Added TypeErasedAbstractOptimizer that automagically implements the interface AbstractOptimizer.
[Shark 3.0] Reworked parts of the algorithms' implementations based on TypedIndividual.
[Shark 3.0] ADDED: MultiVariateNormalDistribution based on EVD of covariance matrix.
[Shark 3.0] ADDED: MultiVariateNormalDistribution based on EVD of covariance matrix.
[Shark 3.0] ADDED: MOCMA, NSGAII, IBEA, SMS-EMOA, PAES
[Shark 3.0 - ErrorFunction] Corrected include-directive for search spaces.
[Shark 3.0] REMOVED: Obsolete header for SearchSpaces.h.
[Shark 3.0] EA component: New algorithms and fitness functions.
[Shark 3.0] ADDED LZ suite of fitness functions. Fixed a bug in the LeastHypervolumeContributorApprox. scheme. Documented the class TypedIndividual.h.
Renamed RNG.cpp (branches/SharkReworkNewInterface/Shark/src/Rng/RNG.cpp) to RNGBackup.cpp.
[Shark 3.0] Checkin started for reworked Fuzzy library.
[Shark 3.0] Adjusted handling of shared pointers.
[Shark 3.0] Const counter initial checkin.
[Shark 3.0] MODIFIED: Fuzzy component. The high level components like RuleBase and Rule still need work.
[Shark 3.0] boost::spirit based parser for the FCL.
Preparations for bugfix release 2.3.3.
[Shark2] Reworked the CMake setup.
[Shark2] REMOVED: Old cmake directory.
[Shark2] FIXED: Cleaned up the whole shark cmake setup.
[Shark2] ADDED: #define NOMINMAX to prevent generation of min and max macros polluting the global namespace on windows.
Created directory 'tags/shark 2.3.3'.
Tagged release 2.3.3 of the shark 2 series.
[Utilities] FIXED: Compilation error for wall-clock timer under Mac OSX.
[Utilities] Fixed preprocessor macros
[Utilities.h] Mac OSX does not support cstdint, switched to stdint.h.
[Shark2] FIXED: rt is not required on osx.
[Shark2] UNIX AND NOT APPLE.
[Shark 2.3.4] Adjusted patch version in CMakeLists.txt
[Shark 3.0] FIXED: Added missing return statements.
[Shark 3.0] ADDED: Abstract classes for multi-objective optimizers and classes.
[Shark 3.0] FIXED: Added missing return statement.
[Shark 3.0] ADDED: Doxygen documentation.
[Shark 3.0] Ported changes to cmake layout over to Shark 3.0. Kept the cmake-folder to ensure a working setup in transitional phases.
[Shark 3.0] FIXED: A problem with missing boost library paths under windows.
[Shark 3.0] ADDED: Building of sphinx-documentation (quite rough yet).
[Shark 3.0] ADDED: Missing include in AbstractBoxConstraintsProvider.
[Shark 3.0] Added missing typenames in AbstractBoxConstraintsProvider.
(Shark 3) ADDED: Abstract charting component for simple visualization tasks, e.g., in unit tests etc.
(Shark 3) MODIFIED: Algorithm AGE I, Benchmark for objecitve functions and mo optimizers.
[Shark 3.0] Small changes, replaced custom constants with boost::math constants.
[Shark 3.0] Added missing includes and fixed template compiler issues to allow for compilation under linux.
[Shark 3] ADDED: FPRAS for approximating the union of boxes.
[Shark 3.0] MODIFIED: Highchart renderer now correctly encodes the chart type.
[Shark 3.0] FIXED: Rely on RngType.
[Shark 2]FIXED: crowdedDistance now calls crowdingDistance internally.
[Shark 3] ADDED: Executable to run SteadyStateMOCMA for cluster experiments.
[Shark 3] ADDED: Executable to run SteadyStateMOCMA for cluster experiments.
[Shark 3] FIXED: Added missing typedefs.
[Shark 3] FIXED: Renamed include file.
[Shark 3] ADDED: Executable for AGE 1.
[Shark 3] ADDED: Executable for AGE 1.
[Shark 3]ADDED: Command line parameter to specify the result directory.
[Shark 3.0] ADDED: Functionality for command line (\mu+1)-MO-CMA.
[Shark 3.0] ADDED: Functionality for command line (\mu+1)-MO-CMA.
[Shark 3.0] FIXED: Numerous problems for large scale experiments on grids.
[Shark 3.0] FIXED: Numerous problems for large scale experiments on grids.
[Shark 3] ADDED: Class for modelling a singleton.
[Shark 3.0] ADDED: Changes for further support of experiments on compute clusters based on GridEngine by Oracle.
[Shark 3.0] FIXED: Compilation problems with gcc.
[Shark 3] FIXED: Problem with missing boost::filesystem_error.
[Shark 3.0] REMOVED: Obsolote CMake configuration.
[Shark 3] REMOVED: Old build system files.
[Shark 3] REMOVED: Old build system files.
[Shark 3]ADDED: Coding Conventions.
[Shark 3.] MODIFIED: Executables are now put into bin and lib.
[Shark 3] ADDED: AGE2
[Shark 3.] REMOVED: Tmp-file for ApproximatedHypervolumeMain.cpp.
[Shark 3.0] MODIFIED: Moved main's to example directory.
[Shark 3.0] MODIFIED: Example executables are compiled from source in example directory now.
[Shark 3.] MODIFIDED: Moved example on calculation of statistics to the examples directory.
[Shark 3.0] MODIFIED: Example on usage of statistics.
[Shark 3] FIXED: Generation of doxygen documentation.
[Shark 3] Switched to static linking of boost libraries.
[Shark 3] MODIFIED: Correct path to conf.py.
[Shark 3] ADDED: Doxygen documentation.
[Shark 3.0] ADDED: Doxygen documentation.
[Shark 3.0] ADDED: Doxygen documentation.
[Shark 3] ADDED: Doxygen documentation.
[Shark 3] ADDED: Doxygen documentation.
[Shark 3.0] A bunch of changes.
[Shark 3.0] ADDED: Tournament and Roulette Wheel Selection.
[Shark 3] A bag of changes - Part II.
[Shark 3] ADDED: Simple compile test for selection operators.
[Shark 3] ADDED: Missing include in AbstractSingleObjectiveOptimizer.h
[Shark 3] FIXED: COmpiler error on mac osx.
[Shark 3] Fix for missing snprintf.
[Shark 3.0] MODIFIED: Library is now static on all platforms.
[Fuzzy] MODIFIED: Fuzzy library is now header-only.
[Fuzzy] MODIFIED: Fuzzy library is now header-only.
[Fuzzy] ADDED: Example illustrating the use of the fuzzy component (not fully functional yet).
[Shark 3] REMOVED: Obsolete classes.
[Shark 3.0] ADDED: Example for generational MOCMA.
[Shark 3.0] ADDED: Tests for single-objective optimizers.
[Shark 3] ADDED: Doxygen documentation for (1+1)-ES.
[CMake Setup] MODIFIED: Boost is now linked in statically by default.
[Shark 3] ADDED: Implementation of name() for TypeErasedSingleObjectiveOptimizer.
[Global RNG] FIXED: Discrete uniform random numbers are now sampled from the correct interval.
[EA - Selection] ADDED: Error handling code.
[Renderer - Highchart] ADDED: Missing include.
[EA - Selection - UnitTest] MODIFIED: Extended testing of the selection methods.
[testFunction - helper] MODIFIED: Refactored the method to rely on boost::parameter.
[Rng] ADDED: Test cases for different distributions.
[Rng] ADDED: Test cases for different distributions.
[Shark - Rng] Commented out the test I accidently checked in.
[Shark 3 - KernelRegression] FIXED: Changed Set to DataContainer.
[Rng Cleanup] Component is now header only, cleaned up association of distributions with underlying generators.
[Rng Cleanup] Component is now header only, cleaned up association of distributions with underlying generators.
[Rng Cleanup] Component is now header only, cleaned up association of distributions with underlying generators.
[Rng Cleanup] Component is now header only, cleaned up association of distributions with underlying generators.
[Shark 3] ADDED: AbstractVectorSpaceObjectiveFunction.h, Factory.h
[Rng] Fix for circular dependencies. Compiles on Win.
[Rng] Fix for circular dependencies. Compiles on Win.
[Rng] Fix for circular dependencies. Compiles on Win.
[Core] Adjustments to interfaces, unification of this_type and base_type missing.
[Core] Adjustments to core interfaces.
[Core] Adjustments to core interfaces.
[Core] Adjustments to core interfaces.
[Core] Adjustments to core interfaces.
[Benchmarks] ADDED: Missing derivativeTestHelper.h.
[Rng] Adjustments for changes to Rng component.
[Shark 3] REMOVED: Compilation of RNG as the component is header only now.
[Shark Defs] REMOVED: custom implementations of erf(c) and gamma and replaced with boost equivalents.
[Rosenbrock] FIXED: Failing unit test due to missing initialization of dimension.
[Csv Test Case] FIXED: Adjustments for renaming of data sets.
[Dirichlet] FIXED: wrong name of gamma function.
[GlobalRng] ADDED: Missing typename.
[Shark 3.0] A bunch of fixes for compilation problems on gcc (see ReviewRequest 30).
[Shark 3 Test] ADDED: Command line arguments for testing environment for integration with build server.
[Shark Test Setup] ADDED: Missing command line argument to switch off reporting.
[Shark 3] REMOVED TEMPORARILY: Rng Distributions test case.
[Shark 3 CMakeSetup] ADDED: Missing link library pthread for linux.
[ObjectiveFunctions/Benchmarks] FIXED: Issue with dimension not being initialized correctly in c'tor.
[Shark 3] ADDED: Parameter fstop in test method.
[Shark 3] Removed custom implementations of erf(c) and gamma and replaced with boost equivalents.
[Shark 3] FIXED: Replaced shark::max with std::max.
[Shark 3] FIXED: min -> std::min
[Shark 3] ADDED: Missing header Core/Exception.h.
[Shark 3] FIXED: shark::max -> std::max.
[Direct Search Algorithms]
[Shark 3] Rename of MatrixType to VectorMatrixType.
[AbstractLoss] ADDED: Missing include.
[AbstractCost] [Shark 3] Rename of MatrixType to VectorMatrixType.
[AbstractCost] [Shark 3] Rename of MatrixType to VectorMatrixTraits.
[Shark 3] Bunch of changes for doing experiments with evolutionary algorithms.
[Shark 3] Bunch of changes for doing experiments with evolutionary algorithms.
[Shark 3] Bunch of changes for doing experiments with evolutionary algorithms.
[Shark 3][MOO Experiment] FIXED: Wrong include and call to templated method.
[Shark 3] Bunch of changes for doing experiments with evolutionary algorithms.
[Shark 3][MOO Experiment] FIXED: Wrong include and call to templated method.
[SMSEMOA] FIXED: Problem in selection for reproduction.
[Shark3] ADDED: Unit test for hypervolume algorithms, both exact and approximated.
[Shark 3] ADDED: Unit test for hypervolume calculation algorithms.
[Shark 3] ADDED: Unit test for hypervolume calculation algorithms.
[Shark 3] ADDED: Unit test for hypervolume calculation algorithms.
[SMSEMOA] FIXED: Problem in selection for reproduction.
[SMSEMOA] FIXED: Problem in selection for reproduction.
[SMSEMOA] FIXED: Issue with wrong initialization of search space bounds and subsequently, relying on an initialized penalty factor.
[SMSEMOA] FIXED: Issue with wrong initialization of search space bounds and subsequently, relying on an initialized penalty factor.
[ObjectiveFunctionTraits] FIXED: Default bounds set to (-)1E20 to allow for computations with bounds.
[InterruptibleAlgorithmRunner] ADDED: Options for querying the known objective functions.
[Shark Test Invocation] MODIFIED: Switched the log level from 'all' to 'test_suite' to reduce size of xml-result files.
[Shark Test Invocation] MODIFIED: Switched the log level from 'all' to 'test_suite' to reduce size of xml-result files.
[Shark 3.] Cleaned up and documented the current CMake setup.
[Shark 3.] Cleaned up and documented the current CMake setup.
[Shark 3.] Cleaned up and documented the current CMake setup.
[DirectSearch - UnitTest] ADDED: Generic test for all known multi-objective optimizers.
[Shark 3.] Cleaned up and documented the current CMake setup.
[Shark 3 CMake setup] FIXED: Added a missing CMAKE_CURRENT_SOURCE_DIR to handle out-of-source builds.
[Direct Search] ADDED: Class for storing Pareto-front approximations.
[Algorithms - DirectSearch] Refactored large parts of the code.
[CVFolds] FIXED: Compiles fine now.
[Algorithms - DirectSearch] Refactored large parts of the code.
[TypedIndividual] FIXED: Removed typename in wrong place.
[InterruptibleAlgorithmRunner] FIXED: Switched to boost::signals1::signal for compatibility with gcc 4.0.x.
[SimulatedBinaryCrossover] ADDED: Missing include.
[[InterruptibleAlgorithmRunner] FIXED: Switched to boost::signals1::signal for compatibility with gcc 4.0.x.
[DirectSearch Algorithms UnitTest] MODIFIED: Lowered maximum running time to 2 minutes.
[Rng UnitTest] MODIFIED: Switched to boost/math/distributions for calculation of mean and variance. Review passed.
[CSvm] FIXED: Added missing header file to example.
[CSvmGridSearch] FIXED: Added missing header file to example.
[GaussianRbfKernel] FIXED: Missing or obsolute includes.
[GaussianRbfKernel] FIXED: Missing or obsolute includes.
[Shark CMakeSetup] ADDED: Static code analysis target for cppcheck on unix.
[Shark CMakeSetup] ADDED: Static code analysis target for cppcheck on unix.
[LinAlg/Base.h] REMOVED: Default implementation of operator== for RealVector.
[Probe Framework] Initial commit, review passed.
[CMake Setup Documentation] Initial reviewed version.
[Shark Logging Facilities] Initial reviewed version.
[Shark Logging Facilities] Initial reviewed version.
[Shark CMake] ADDED: Link against boost::date_time.
[Probe] FIXED: Compilation Problem on Linux.
[Probe] FIXED: Compilation Problem on Linux.
[Probe] FIXED: Compilation Problem on Linux.
[Probe] FIXED: Compilation Problem on Linux.
[Probe] FIXED: Compilation Problem on Linux.
[Probe] FIXED: Compilation Problem on Linux(internal compiler error: in add_phi_arg, at tree-phinodes.c:391).
[Probe] FIXED: Compilation Problem on Linux(internal compiler error: in add_phi_arg, at tree-phinodes.c:391).
Shark CMake Setup: ADDED: Missing boost library regex.
[Shark Benchmark Functions] ADDED: Himmelblau function, passed review.
[Least Contributor Approximator] Small changes to prepare for pure C bindings (Matlab etc.).
[InterruptibleAlgorithmRunner] Switched to logger-facilities for reporting.
[LinAlg] Preparation for inclusion of GPGPU in shark. Reviewed.
[LinAlg] Preparation for inclusion of GPGPU in shark. Reviewed.
[LinAlg] Preparation for inclusion of GPGPU in shark. Reviewed.
[LinAlg/Base.h] Reverted to previous state.
(LinAlg) FIXED: Problem with concatenation of macro arguments in for gcc.
[include/Core] ADDED: svn:ignore for Shark.h.
[Shark CMakeSetup] Forced libtool-usage for generating static shark library under OSX.
[Shark DirectSearch] Small changes in preparation for examples.
[Shark 3] Preparations for inclusion of EA tutorials.
[Shark DirectSearch] First bunch of EA examples/tutorials.
[Shark Documentation] FIXED: Bunch of oxygen errors originating in DirectSearch.
[Shark Documentation] FIXED: Bunch of doxygen errors and issues.
[Shark Documentation] REMOVED: Obsolete Fuzzy includes.
[Shark Documentation] REMOVED: Obsolete impl for Statistics.
[Shark Cleanup] REMOVED: Obsolete headers.
[Shark DirectSearch] ADDED: Missing \brief documentation.
[Shark] REMOVED: UF2, Welding, Weldbead.
[Shark UnitTest Reporting] Switched log level to test_suite again.
[Shark3] ADDED: Tutorials and documentation for direct search in general and for CMA/MOCMA.
[PGM] FIXED: Compilation error with access to directory iterator.
[CMA Tutorial] ADDED: Description of CMA Probes tutorial, reference to literature.
[General documentation] ADDED: First proposal for a review process for external contributors.
[Getting Help] ADDED: Bug reporting guidelines and description of available support channels.
[CMake Setup] ADDED: Option OPT_DYNAMIC_LIBRARY that controls whether a dynamic library or a static library of the shark library is built.
[Tutorials] ADDED: Tutorial on factory-method-pattern implementation.
[Documentation] ADDED: Contents for getting help.
[Documentation] ADDED: Example for factory-usage.
[Documentation] ADDED: Very brief tutorial on shark::Statistics.
[Hypervolume Approximation] ADDED: Handling of configuration parameters for hypervolume-based selection.
[Shark DirectSearch] ADDED: Selection and recombination operators for single-objective optimization.
[Shark DirectSearch/Selection] MODIFIED: Temporarily marked EPTournament-Selection test as failing.
[Shark NegativeGaussianProcessEvidence] MODIFIED: Adjusted threshold in unit test.
[Shark] REMOVED: Obsolete class ResultSet.
[Shark] ADDED: Missing brief documentation.
[cpp-netlib] Initial checkin.
[Shark MOO Examples] ADDED: Command line bounding box calculator
[Shark MOO Examples] ADDED: Command line bounding box calculator
[Shark MOO Examples] ADDED: Command line bounding box calculator
[Shark] MODIFIED: Calculation of median in statistics component.
[Shark Network] ADDED: Networking unit tests and networking example based on extjs.
[Shark Doc Installation] MODIFIED: Installation procedure for documentation.
FIXED: Problems with out-of-source builds.
[Shark Sphinx Setup] MODIFIED: Renamed sphinxcontrib.doxylink to doxylink.
[Shark Doxygen Setup] ADDED: Missing configure directive in Shark.dox.in.
[Shark Doxygen Setup] FIXED: Missing "/".
[Shark Sphinx Setup] MODIFIED: Retrieval of doxylink extension.
[Shark Doxygen Installation Setup] ADDED: Installation of css files.
[Shark Network Example] ADDED: Installation of supplementary files.
[Shark Benchmark Functions] FIXED: Numerical instabilities in DTLZ3 and DTLZ4.
[Shark Network Unittest] MODIFIED: Number of requests lowered.
FIXED: Accidently wrote out addresses instead of version numbers.
[Shark Installation] FIXED: Installation of header files.
[Shark Direct Search] ADDED: Explicit bitflip mutation.
[Shark] FIXED: Issues with boost 1.47
[Shark Data] MODIFIED: Adjustments for switch to boost filesystem ver. 3
[Shark Documentation] MODIFIED: Moved configure macros for documentation setup.
[Shark doc] ADDED: Documentation for missing cmake options.
FIXED: std::pow issue on MS Windows.
[Shark] Tag before commit of new include structure.
[Shark] Tag before commit of new include structure.
[Shark] MODIFIED: Removed installation of unit tests.
[Shark] MODIFIED: Installation of header files to include/shark/... .
[Shark] FIXED: Problem in factory tutorial.
[Shark XNES] \cond-out detail namespace.
[Shark] MODIFIED: Correct naming of packages according to platform.
Accounted for missing \\brief
Accounted for missing brief.
Removed references to RobustParetoDominanceComparitor.
[Shark] FIXED: Installation of data for supervised tutorials.
[Shark] FIXED: Compilation problem on windows.
[Shark] FIXED: Compilation problem on windows.
[Shark Deb Setup] FIXED: Problems with debian package installation.
[Shark/Direct Search] ADDED: Access methods for covariance matrix, sigma etc..
[Fuzzy] First bunch of fixes and new unit tests.
[Fuzzy] ADDED: Missing files for Fuzzy Control Language Parser.
[CMake setup] ADDED: Dependency on boost::utf.
[Fuzzy] MODIFIED: Disabled fuzzy unit test for the moment.
[Fuzzy] FIXED: Compilation problems with boost 1.44.
[Fuzzy] ADDED: Factories for fuzzy sets and linguistic terms. ADDED: Set of units tests for the overall component and the fuzzy control language (FCL) parser.
ADDED: debian subdirectory to prepare for packaging for debian and ubuntu.
[Debian packaging] MODIFIED: Copyright and description of package.<
[CMake setup/Debian package] Added automatic build and installation of doxylink.
REMOVED: Obsolete include files for fuzzy control language parser.
FIXED: Warning of unused parameters.
FIXED: Path to FCL parser files.
Fixes for compile time warning with boost::serialization. Only 300 more
Fixed a bunch of compile time warnings w.r.t. boost::serialization.
Added missing nvp generation for serialization.
Added example for Ackley fitness function.
Removed dependency on target doxylink.
Fixed an issue with the overmars-yap hanging in certain corner-cases.
Adjusted serialization to generate name-value-pairs.
ADDED: method searchPoint() in TypedIndividual to access search point more verbosely.
MODIFIED: Moved optimizer traits to core module.
Fixed a build error on Windows.
Added a JSON output formatting for Pareto-front approximations to ease integration of MOO experiment results with other systems/programming languages. Introduced a boost::spirit-based parser and generator for json to the Data-module.
Fixed an error reported by the PathScale compiler.
Introduced log-normal distributed numbers to GlobalRng.
Fix warnings in testFunction.
Fix build issues with clan/llvm 3.0.
Removed OpenMP parallel for loop for sampling purposes.
Bumped boost version to 1.48.
Fixed naming of include guard, Fixed naming of methods.
Fix an issue with wrongly reported box-constraints of objective functions.
tuantx711 (2):
mc-svm online solver code
commit f5dc2794d3d208abd74f5f00d90486dc47885e8c
unknown (2):
fixed a quadrillion of warnings on vc12 64Bit
improved VC12 support by adding project and file groups
-----------------------------------------------------------------------
No new revisions were added by this update.
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/debian-science/packages/shark.git
More information about the debian-science-commits
mailing list