|
|
c8e73cd |
1.04: 2000/6/17, add "load" button to svm-toy. README
|
|
|
c8e73cd |
file updated.
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.0: 2000/8, major updates. Include nu-svm, one-class svm, and svr
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.01: 2000/9/22, correct the calculation of obj value and number of bounded support vectors
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.02: 2000/9/29, replace b^2/2 newsvm to regular nu svm.
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.03: 2000/10/24 some improvements on the computational speed
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.1: 2000/12/19 Java version included, regression demonstrated in svm-toy
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.2: 2001/1/16 multi-class classification, nu-SVR, remove epsilon_a
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.3: 2001/3/15 c+-, cross validation, fix some minor bugs
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.31: 2001/4/12 fix one bug on one-class SVM, use float for Cache
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.32: 2001/9/23
|
|
|
c8e73cd |
1. max line number of svm-scale now dynamic
|
|
|
c8e73cd |
2. gcc 3.0 problem: now g++ always used
|
|
|
c8e73cd |
3. java code in a "libsvm" package
|
|
|
c8e73cd |
4. avoid a problem when compiled with Borland C++ builder
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.33: 2001/12/1
|
|
|
c8e73cd |
Python interface added
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.34: 2002/6/15
|
|
|
c8e73cd |
Add the subroutine svm_check_parameter in svm.cpp
|
|
|
c8e73cd |
--> better error handling
|
|
|
c8e73cd |
fix bug of python interface for handling different weights
|
|
|
c8e73cd |
fix bug of cross validation in svm-train.c
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.35: 2002/6/16
|
|
|
c8e73cd |
libsvm.jar was not compiled correctly in 2.34
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.36: 2002/8/4
|
|
|
c8e73cd |
grid.py added: contour plot of CV accuracy
|
|
|
c8e73cd |
fix several bugs
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.4: 2003/4/1
|
|
|
c8e73cd |
svm.cpp
|
|
|
c8e73cd |
non-psd kernel using max(...,0) in svm.cpp
|
|
|
c8e73cd |
python interface
|
|
|
c8e73cd |
python interface bug (nu-svm)
|
|
|
c8e73cd |
grid.py
|
|
|
c8e73cd |
-log2c and -log2c for grid.py
|
|
|
c8e73cd |
output current best
|
|
|
c8e73cd |
coarse grid as default (2)
|
|
|
c8e73cd |
ssh for grid.py
|
|
|
c8e73cd |
improvements of scaling
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.5: 2003/11/13
|
|
|
c8e73cd |
subroutines for accessing decision values and number of labels.
|
|
|
c8e73cd |
for svm.cpp, java, and python interface
|
|
|
c8e73cd |
fix bug of svm-scale.c (about -r and -s factors)
|
|
|
c8e73cd |
use fscanf but not sscanf in svm-predict.c (faster)
|
|
|
c8e73cd |
makefile for windows
|
|
|
c8e73cd |
add "using namespace std;" in some .java files
|
|
|
c8e73cd |
improve easy.py: output cv rate, error messages printed
|
|
|
c8e73cd |
better checking
|
|
|
c8e73cd |
better python interface example svm_test.py
|
|
|
c8e73cd |
and some minor updates
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.6: 2004/04/01
|
|
|
c8e73cd |
Probability estimates for classification/regression
|
|
|
c8e73cd |
Python interface: use swig 1.3 instead of 1.1
|
|
|
c8e73cd |
Cross validation becomes a library subroutine
|
|
|
c8e73cd |
A few minor corrections: (not completely listed)
|
|
|
c8e73cd |
more interface functions such as getting svm_type
|
|
|
c8e73cd |
print nu only when Cp=Cn
|
|
|
c8e73cd |
floor division in python interface
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.7: 2004/11/10
|
|
|
c8e73cd |
Stratified cross validation
|
|
|
c8e73cd |
Better faq
|
|
|
c8e73cd |
Scaling: support storing the factor of y
|
|
|
c8e73cd |
A few minor updates:
|
|
|
c8e73cd |
class QMatrix added
|
|
|
c8e73cd |
improve the use of easy.py and grid.py on windows
|
|
|
c8e73cd |
grid.py: same CV and same g: use smaller C
|
|
|
c8e73cd |
sparse input for python interface
|
|
|
c8e73cd |
working set selection: < to <=
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.71: 2004/11/20
|
|
|
c8e73cd |
fix a java bug introduced from 2.6 to 2.7
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.8: 2005/04/01
|
|
|
c8e73cd |
new working set selection via second order information
|
|
|
c8e73cd |
fix minor changes/corrections:
|
|
|
c8e73cd |
problem when cache size less than two kernel columns
|
|
|
c8e73cd |
-v #data -> stratified CV is not loo -> ensure loo is done
|
|
|
c8e73cd |
problem of typing "svm-train -c" only
|
|
|
c8e73cd |
problem of "svm-train -n 1 -s 3 ..."
|
|
|
c8e73cd |
python interface makefile: -fPIC for 32/64bit
|
|
|
c8e73cd |
color change in svmtoy
|
|
|
c8e73cd |
makefile in building QT svmtoy
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.81: 2005/11/20
|
|
|
c8e73cd |
add a python script subset.py for subsampling
|
|
|
c8e73cd |
slightly modify the working set so it's exact the same as the paper
|
|
|
c8e73cd |
default cache size to 100 MB
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.82: 2006/04/01
|
|
|
c8e73cd |
precomputed kernel
|
|
|
c8e73cd |
directly implement a fast powi() function
|
|
|
c8e73cd |
poly degree double to int
|
|
|
c8e73cd |
minor corrections:
|
|
|
c8e73cd |
java code for CV
|
|
|
c8e73cd |
2.83: 2006/11/17
|
|
|
c8e73cd |
Fix the bug of -t 3
|
|
|
c8e73cd |
better checking load/save in svm-train.c/svm-predict.c/svm.cpp
|
|
|
c8e73cd |
remove redundant var pos in svm_predict_values (thanks to Albert Strasheim)
|
|
|
c8e73cd |
Better descriptions in README for the precomputed kernel
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.84: 2007/04/01
|
|
|
c8e73cd |
Improve the shrinking code: faster for some cases
|
|
|
c8e73cd |
Code more 64-bit friendly: allow large -m
|
|
|
c8e73cd |
In Solver, b is replaced by p
|
|
|
c8e73cd |
Subroutine max_violating_pair removed. Things are directly
|
|
|
c8e73cd |
written in do_shrinking().
|
|
|
c8e73cd |
Modify do_shrinking() so variable names are the same as libsvm
|
|
|
c8e73cd |
document
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.85: 2007/11/6
|
|
|
c8e73cd |
fix minor memory leak in svm-predict.c
|
|
|
c8e73cd |
add tools/checkdata.py
|
|
|
c8e73cd |
java to 1.5
|
|
|
c8e73cd |
Makefile: CXX?
|
|
|
c8e73cd |
Makefile.win: avoid warning from .net 2005
|
|
|
c8e73cd |
avoid warning of gcc 4.2
|
|
|
c8e73cd |
sigma 0.001 to 1e-12; in Newton direction of prob output
|
|
|
c8e73cd |
2 * (long int) l avoid warning of old Visual C++
|
|
|
c8e73cd |
grid.py: avoid gnuplot problem on windows
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.86: 2008/04/01
|
|
|
c8e73cd |
svm-scale for java
|
|
|
c8e73cd |
version number in svm.h and svm.m4
|
|
|
c8e73cd |
rename svmtrain.exe to svm-train.exe
|
|
|
c8e73cd |
python: while 1 --> while True, Popen -> call
|
|
|
c8e73cd |
show best parameters on the contour of grid.py
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.87: 2008/10/13
|
|
|
c8e73cd |
svm-toy/qt updated to qt4 from qt3
|
|
|
c8e73cd |
fix a bug in svm-scale.c
|
|
|
c8e73cd |
max feature index of -r file is considered
|
|
|
c8e73cd |
Makefile: add make lib; add -Wconversion and -fPIC in Makefile
|
|
|
c8e73cd |
Add "rb" in load_model of svm.cpp
|
|
|
c8e73cd |
Simplify do_shrinking of svm.cpp
|
|
|
c8e73cd |
Change the order of loops in reconstrict_gradient of svm.cpp
|
|
|
c8e73cd |
save the number of kernel evaluations
|
|
|
c8e73cd |
Add python/setup.py
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.88: 2008/10/30
|
|
|
c8e73cd |
better gradient reconstructions
|
|
|
c8e73cd |
issue a warning when -h 0 may be faster
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.89: 2009/04/01
|
|
|
c8e73cd |
reduce input/loading time of svm-train/svm-predict by half
|
|
|
c8e73cd |
pointer function so users can specify their own outputs
|
|
|
c8e73cd |
remove info_flush()
|
|
|
c8e73cd |
a extern variable libsvm_version
|
|
|
c8e73cd |
svm-train -q option (disable outputs)
|
|
|
c8e73cd |
svm-scale: warning if more nonzero produced
|
|
|
c8e73cd |
easy.py: popel.communiate() to avoid some deadlock (if lots of
|
|
|
c8e73cd |
outputs when #classes is large)
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.9: 2009/11/1
|
|
|
c8e73cd |
tools/*.py can be run under python 3.0
|
|
|
c8e73cd |
svm_set_quiet() in python interface to disable outputs
|
|
|
c8e73cd |
check gamma < 0
|
|
|
c8e73cd |
internal functions to be static
|
|
|
c8e73cd |
|
|
|
c8e73cd |
2.91: 2010/04/01
|
|
|
c8e73cd |
completely new python interface using ctype
|
|
|
c8e73cd |
new way to set the print_string function
|
|
|
c8e73cd |
Java: able to load model from a BufferedReader directly
|
|
|
c8e73cd |
fix grid.py so -log2c can be run under python 2.6 or after
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.0: 2010/09/13
|
|
|
c8e73cd |
Move model structure to svm.h
|
|
|
c8e73cd |
Two functions for freeing a model (content or the whole model)
|
|
|
c8e73cd |
QD from Qfloat to double (better precision because SSE on 64-bit machines less accurate than i387 on 32-bit
|
|
|
c8e73cd |
exit status for checkdata.py
|
|
|
c8e73cd |
old python interface (swig) is removed
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.1: 2011/04/01
|
|
|
c8e73cd |
MATLAB interface:
|
|
|
c8e73cd |
Merge matlab interface to core libsvm
|
|
|
c8e73cd |
Using mexPrintf() when calling info() in MATLAB interface.
|
|
|
c8e73cd |
Both 32- and 64-bit windows binary files are provided
|
|
|
c8e73cd |
Java:
|
|
|
c8e73cd |
Math.random is replaced by Random in java interface
|
|
|
c8e73cd |
Python interface:
|
|
|
c8e73cd |
i subroutines to get SVs
|
|
|
c8e73cd |
relative path to load *.dll and *.so
|
|
|
c8e73cd |
svm.cpp:
|
|
|
c8e73cd |
null pointer check before release memory in svm_free_model_content()
|
|
|
c8e73cd |
svm_destroy_model() no longer supported.
|
|
|
c8e73cd |
svm-train.c and svm-predict.c
|
|
|
c8e73cd |
Better format check in reading data labels
|
|
|
c8e73cd |
svm-toy:
|
|
|
c8e73cd |
fix the svm_toy dialog path
|
|
|
c8e73cd |
tools:
|
|
|
c8e73cd |
Using new string formatting/encoding in tools/*.py
|
|
|
c8e73cd |
clearer png output, fix grid.py legend
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.11: 2011/11/5
|
|
|
c8e73cd |
Set max number of iterations in the main loop of solvers
|
|
|
c8e73cd |
matlab:
|
|
|
c8e73cd |
new make.m for unix/mac/windows and for matlab/octave
|
|
|
c8e73cd |
matlab and python:
|
|
|
c8e73cd |
fix a problem that decision values returned by svmpredict is empty if number of classes = 1
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.12: 2012/04/01
|
|
|
c8e73cd |
svm-toy: support loading/saving of regression data
|
|
|
c8e73cd |
python interface: handle the issue of the "0th" feature if using lists
|
|
|
c8e73cd |
tools/grid.py: not redrawing contour if c,g, cv doesn't change
|
|
|
c8e73cd |
add setlocale when saving and loading model so not affected by users' locale
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.13: 2012/11/06
|
|
|
c8e73cd |
svm.c and svm.h:
|
|
|
c8e73cd |
add sv_indices in model structure, so users can know which training instances are SVs
|
|
|
c8e73cd |
two library funs svm_get_sv_indices and svm_get_nr_sv are added
|
|
|
c8e73cd |
max_iter warning moved to stderr so -q won't disable it
|
|
|
c8e73cd |
svm-train.c:
|
|
|
c8e73cd |
usage modified to stress that multiclass is supported
|
|
|
c8e73cd |
svm-predict.c:
|
|
|
c8e73cd |
add -q for svm-predict
|
|
|
c8e73cd |
svm-scale.c:
|
|
|
c8e73cd |
issue a warning if feature indices do not start from 1
|
|
|
c8e73cd |
issue a warning for inconsistency between scaling-factor file and input file
|
|
|
c8e73cd |
tools:
|
|
|
c8e73cd |
subset.py is written to be much faster
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.14: 2012/11/16
|
|
|
c8e73cd |
fix a bug: we didn't re-compile windows binary after updating svm.def
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.15: 2013/1/27
|
|
|
c8e73cd |
fix the bug of not freeing sv_indices
|
|
|
c8e73cd |
tools/grid.py:
|
|
|
c8e73cd |
-null option: allow the search on C or g only
|
|
|
c8e73cd |
-resume option: resume tasks from an earlier run
|
|
|
c8e73cd |
can be called as a python module
|
|
|
c8e73cd |
python interface:
|
|
|
c8e73cd |
local package searched first
|
|
|
c8e73cd |
libsvm options can be str or list
|
|
|
c8e73cd |
param.show() becomes print(param)
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.16: 2013/1/27
|
|
|
c8e73cd |
tools/:
|
|
|
c8e73cd |
easy.py fails in 3.15. Fix it by modifying grid.py
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.17: 2013/04/01
|
|
|
c8e73cd |
svm.cpp:
|
|
|
c8e73cd |
if class labels are 1 and -1, ensure labels[0] = 1 and labels[1] = -1
|
|
|
c8e73cd |
initialize model->sv_indices as null in svm_load_model
|
|
|
c8e73cd |
if nr_fold > # data, change nr_fold to be # data and ro leave-one-out cv
|
|
|
c8e73cd |
matlab interface:
|
|
|
c8e73cd |
handle the problem where output variables are not specified
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.18: 2014/04/01
|
|
|
c8e73cd |
svm.cpp and svm-scale.c:
|
|
|
c8e73cd |
check return values of fscanf
|
|
|
c8e73cd |
matlab interface:
|
|
|
c8e73cd |
Makefile no longer handles octave because make.m should be used
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.19: 2014/10/27
|
|
|
c8e73cd |
svm.cpp:
|
|
|
c8e73cd |
use size_t rather than long int in some places for 64-bit windows support
|
|
|
c8e73cd |
matlab interface:
|
|
|
c8e73cd |
use size_t and mwIndex to avoid type conversions
|
|
|
c8e73cd |
python interface:
|
|
|
c8e73cd |
use __all__
|
|
|
c8e73cd |
|
|
|
c8e73cd |
3.20: 2014/11/15
|
|
|
c8e73cd |
matlab interface:
|
|
|
c8e73cd |
fix a bug in libsvmwrite.c introduced in 3.19 (cannot run on windows)
|
|
|
f7acaea |
|
|
|
f7acaea |
3.21: 2015/12/14
|
|
|
f7acaea |
pre-built windows exe files changed from 32 to 64 bit
|
|
|
f7acaea |
matlab interface:
|
|
|
f7acaea |
now use #include "../svm.h"
|
|
|
e6b06f5 |
fix some minor issues in make.m of matlab interface
|
|
|
e6b06f5 |
|
|
|
e6b06f5 |
3.22: 2016/12/22
|
|
|
e6b06f5 |
probability output:
|
|
|
e6b06f5 |
if 2 classes, directly output the predited probabilities
|
|
|
e6b06f5 |
rather than run the iterative algorithms for multi-class situations
|
|
|
e6b06f5 |
java:
|
|
|
e6b06f5 |
libsvm.jar generated by java 1.7 rather than 1.5
|
|
|
e6b06f5 |
change the use of "_" in svm.java, which won't be allowed in later java
|
|
|
e6b06f5 |
|
|
|
e6b06f5 |
3.23: 2018/7/15
|
|
|
e6b06f5 |
add more digits of predicted file, model file, scaled data and data from matlab libsvmwrite: to %.17g
|
|
|
e6b06f5 |
revise svm-scale.c so features in test data that do not appear in training data are scaled to zero.
|
|
|
e6b06f5 |
remove unnecessary tab or space in all files
|
|
|
e6b06f5 |
python interface:
|
|
|
e6b06f5 |
add Scipy support
|
|
|
e6b06f5 |
add scaling (csr_find_scale_param and csr_scale functions)
|
|
|
e6b06f5 |
put some utility functions identical in LIBLINEAR to commonutil.py.
|
|
|
e6b06f5 |
functions for scaling are put in commonutil.py.
|
|
|
e6b06f5 |
sort column indices of csr matrix before training as feature indices must be ascending
|
|
|
e6b06f5 |
convert Qt version of svm-toy from Qt4 to Qt5
|
|
|
e6b06f5 |
remove gtk svm-toy because we stop maintaining this tool
|
|
|
e6b06f5 |
minor improvement of descriptions in README
|
|
|
e6b06f5 |
|