]>
granicus.if.org Git - liblinear/log
yong [Fri, 24 Oct 2014 11:57:06 +0000 (19:57 +0800)]
Compile windows binary.
yong [Fri, 24 Oct 2014 11:09:48 +0000 (19:09 +0800)]
add three functions:
get_decfun_coef, get_decfun_bias, and check_regression_model
to linear.def
Yu-Chin [Tue, 21 Oct 2014 19:32:51 +0000 (03:32 +0800)]
In get_decfun_coef and get_decfun_bias, a zero value is returned in all invalid
cases. We decide not to return NaN because we are not sure if it can be
properly assigned in all platforms. (For example, ``double nan = 0.0/0.0;''
won't compile in Visual Studio.)
b99902019 [Fri, 17 Oct 2014 03:19:07 +0000 (11:19 +0800)]
To support nnz > 2^32, change the data type of the variables related to nnz.
Data types of some other variables are also changed to be consistent with the
data types returned by matlab functions.
Yu-Chin [Thu, 16 Oct 2014 13:41:23 +0000 (21:41 +0800)]
remove ^M in the end of each line in make.m
Yu-Chin [Wed, 15 Oct 2014 15:17:41 +0000 (23:17 +0800)]
1. use "extern C" in blas to make sure the compiled function names are always not mangled.
2. replace ``#'' with ``='' in matlab/Makefile
Yu-Chin [Mon, 13 Oct 2014 16:13:36 +0000 (00:13 +0800)]
replace long int with size_t in linear.cpp and train.c
Hsiang-Fu Yu [Mon, 28 Jul 2014 18:22:42 +0000 (13:22 -0500)]
add two functions get_decfun_coef and get_decfun_bias
Yu-Chin [Fri, 25 Jul 2014 15:37:50 +0000 (23:37 +0800)]
use __all__ in Python interfaces to make sure only useful things are visible to the users
coha [Thu, 16 Oct 2014 05:43:09 +0000 (13:43 +0800)]
Expand the expression of blas to fix the compiling error
Chih-Jen Lin [Fri, 3 Oct 2014 04:18:08 +0000 (12:18 +0800)]
Merge branch 'master' of svm.csie.ntu.edu.tw:liblr
Chih-Jen Lin [Fri, 3 Oct 2014 04:17:34 +0000 (12:17 +0800)]
stopping condition of l1 solvers uses 1-norm rather than inf norm
(wrong description in README)
Yu-Chin [Fri, 29 Aug 2014 08:51:23 +0000 (16:51 +0800)]
initialize Gnorm1_init with -1
Yu-Chin [Tue, 26 Aug 2014 03:46:27 +0000 (11:46 +0800)]
initialize Gnorm1_init with NAN
Chih-Jen Lin [Fri, 20 Jun 2014 14:15:39 +0000 (22:15 +0800)]
Merge branch 'master' of svm.csie.ntu.edu.tw:liblr
Chih-Jen Lin [Fri, 20 Jun 2014 14:15:19 +0000 (22:15 +0800)]
fix the following bug
- model_->bias = (int)ptr[0];
+ model_->bias = ptr[0];
Tzu-Ming Kuo [Fri, 6 Jun 2014 15:42:35 +0000 (23:42 +0800)]
modify matlab/train.c to check if the label vector is in dense format.
yong [Wed, 23 Apr 2014 15:23:58 +0000 (23:23 +0800)]
Change the year of COPYRIGHT from 2013 to 2014
yong [Fri, 21 Mar 2014 14:54:24 +0000 (22:54 +0800)]
Modify SHVER from 1 to 2.
boyu [Tue, 10 Dec 2013 14:36:59 +0000 (22:36 +0800)]
matlab Makefile modified, octave removed
b99902019 [Mon, 2 Dec 2013 04:32:14 +0000 (12:32 +0800)]
Modify README so that user can build 64 bit windows binary
Fix a bug in Makefile.win: __WIN32__ should be _WIN32
Chih-Jen Lin [Mon, 11 Nov 2013 10:12:19 +0000 (18:12 +0800)]
remove one space in matlab/README
boyu [Sat, 9 Nov 2013 10:21:43 +0000 (18:21 +0800)]
compiled windows binary
Wei-Sheng Chin [Sat, 2 Nov 2013 06:45:49 +0000 (14:45 +0800)]
Replace long with long int in linear.cpp for consistency
Wei-Sheng Chin [Thu, 31 Oct 2013 10:09:08 +0000 (18:09 +0800)]
Change the type of nnz (counter of nonzero elements) in "transpose" in "linear.cpp" from "int" to "long."
Yu-Chin [Sat, 26 Oct 2013 06:19:48 +0000 (14:19 +0800)]
revert previous two accidental commits
Yu-Chin [Sat, 26 Oct 2013 06:00:32 +0000 (14:00 +0800)]
set number of threads
Yu-Chin [Fri, 25 Oct 2013 15:22:32 +0000 (23:22 +0800)]
.
Yu-Chin [Sat, 31 Aug 2013 07:43:06 +0000 (15:43 +0800)]
Modify python README so that users can run the examples without errors.
Yu-Chin [Fri, 16 Aug 2013 06:54:03 +0000 (14:54 +0800)]
fix a bug: use "delete[] C" instead of "delete C" in train_one
Yu-Chin [Wed, 24 Jul 2013 16:40:41 +0000 (00:40 +0800)]
Fix the mixing tab and space problem in tron.cpp.
Tzu-Ming Kuo [Mon, 1 Jul 2013 15:57:43 +0000 (23:57 +0800)]
Modify function group_classes in linear.cpp to ensure label[0] = 1 if class labels are +1 and -1
yong [Thu, 13 Jun 2013 12:23:08 +0000 (20:23 +0800)]
Fix the bug of type error when calling set_to_default_values() in python interface
Chih-Jen Lin [Thu, 16 May 2013 11:01:51 +0000 (19:01 +0800)]
Chun-Heng Huang [Mon, 11 Mar 2013 09:35:30 +0000 (17:35 +0800)]
Updated a switch statement.
Changed it to the convention.
Chun-Heng Huang [Mon, 11 Mar 2013 04:20:04 +0000 (12:20 +0800)]
Matlab interface parameter checking
Check 'nlhs' argument.
Allow various output size for predict.
leepei [Sun, 10 Feb 2013 17:39:46 +0000 (01:39 +0800)]
Let nr_fold = l when nr_fold > l in CV
Signed-off-by: Po-Wei Wang <xflash96@gmail.com>
cjlin [Sun, 27 Jan 2013 08:11:31 +0000 (08:11 +0000)]
change the year to 2013
guestwalk [Fri, 25 Jan 2013 06:58:44 +0000 (06:58 +0000)]
Compile windows binary.
guestwalk [Mon, 21 Jan 2013 18:03:19 +0000 (18:03 +0000)]
Modify Python interface, major changes are:
+ Find local library first.
+ Change show() to __str__.
guestwalk [Fri, 23 Nov 2012 02:48:45 +0000 (02:48 +0000)]
Add return 0 into print_null in predict.c,
because it is used to replace printf for the "-q" option
and printf will return the number of characters it printed.
In train we do not need this because the two "info" are different.
leepei [Sun, 28 Oct 2012 16:46:32 +0000 (16:46 +0000)]
Fix some indention
cjlin [Fri, 19 Oct 2012 00:51:35 +0000 (00:51 +0000)]
fix a typo in make.m
guestwalk [Sat, 13 Oct 2012 12:28:01 +0000 (12:28 +0000)]
fix some typo in README and matlab/make.m
guestwalk [Wed, 10 Oct 2012 08:18:37 +0000 (08:18 +0000)]
update windows binary to correct version. (previous revision is 64-bit)
guestwalk [Wed, 10 Oct 2012 04:49:54 +0000 (04:49 +0000)]
remove duplicated blas.h in blas/Makefile
fix a syntax bug in matlab/predict.c
update windows binary
guestwalk [Wed, 12 Sep 2012 01:00:26 +0000 (01:00 +0000)]
Fix a typo(CXXC) in Makefile.win.
leepei [Wed, 25 Jul 2012 06:25:21 +0000 (06:25 +0000)]
Fixed a bug of -q option of predict in matlab interface
leepei [Fri, 20 Jul 2012 07:59:36 +0000 (07:59 +0000)]
Added -q option for predict and modified the solver type usage message
popo [Sat, 30 Jun 2012 01:15:11 +0000 (01:15 +0000)]
Remove matlab/run.m in trunk.
coha [Tue, 1 May 2012 08:19:50 +0000 (08:19 +0000)]
Modify the make.m of matlab,
the line after else should be moved right.
popo [Thu, 26 Apr 2012 02:57:08 +0000 (02:57 +0000)]
fix typos in python/README.
popo [Wed, 25 Apr 2012 05:32:21 +0000 (05:32 +0000)]
Fix bug. get_label() returns integer list. model.get_type() is not supported.
popo [Fri, 20 Apr 2012 07:48:46 +0000 (07:48 +0000)]
Fix bugs.
- svm_read_problem should accept regression problems.
- The model name in load_model should be encoded.
cjlin [Tue, 17 Apr 2012 05:45:51 +0000 (05:45 +0000)]
copyright year 2012
popo [Tue, 17 Apr 2012 05:31:42 +0000 (05:31 +0000)]
Add windows binary files.
cjlin [Tue, 17 Apr 2012 05:26:07 +0000 (05:26 +0000)]
indicating -b 1 is only for testing
popo [Tue, 17 Apr 2012 04:50:50 +0000 (04:50 +0000)]
Add -s 11, 12, 13 to python and matlab interfaces.
popo [Sun, 15 Apr 2012 06:07:46 +0000 (06:07 +0000)]
- Add regression solvers (-s 11, 12, 13).
- In L1-regularized solvers, Gp <= H*w[j] and Gn >= H*w[j] become
Gp < H*w[j] and Gp > H*w[j].
- For tron solvers, C array is used instead of Cp and Cn as the
argument of the constructors of function classes.
rofu [Wed, 11 Apr 2012 18:38:27 +0000 (18:38 +0000)]
Fix the type "-W1" to the correct option "-Wl"
popo [Tue, 10 Apr 2012 04:28:19 +0000 (04:28 +0000)]
Because model_ is a global variable, do_predict(input, output, model_)
does not need the third argument.
popo [Mon, 9 Apr 2012 14:34:11 +0000 (14:34 +0000)]
In python interface, file names should be encoded in save_model and load_model.
After python 3.2, we need to use bytes instead of a unicode string.
popo [Thu, 22 Mar 2012 03:27:37 +0000 (03:27 +0000)]
In the README file of python interface, p_val should be p_label
in the usage of evaluations().
rofu [Mon, 2 Jan 2012 04:49:38 +0000 (04:49 +0000)]
Makefile is modified such that creating the shared library
works well on unix platforms by looking at the result of uname(1).
- We need this because Mac OS now does not ignore "-soname" which is used
in Linux and BSD.
- We use uname(1) rather than $OSTYPE because uname(1) is available everywhere
- Instead of using the if-else of make, we use the if-else of shell. The
reason is the syntax of if-else for gnu-make and bsd-make are different.
popo [Mon, 26 Dec 2011 08:20:39 +0000 (08:20 +0000)]
Initial w_size according to w in solve_l1r_lr.
locale should be reset in load_model if the model is invalid.
popo [Fri, 9 Dec 2011 11:12:37 +0000 (11:12 +0000)]
Add variable primal_solver_tol for primal solvers, where
primal_solver_tol = eps*max(min(pos,neg), 1)/prob->l.
popo [Tue, 29 Nov 2011 05:19:43 +0000 (05:19 +0000)]
Add setlocale(LC_ALL, "C") when saving and loading model.
cjlin [Tue, 11 Oct 2011 07:01:35 +0000 (07:01 +0000)]
indicate in predict option that -b 1 is for logistic regression only
coha [Thu, 1 Sep 2011 15:43:53 +0000 (15:43 +0000)]
Using try and catch method to show recommend message of make.m
Modify README for new MATLAB make.m
coha [Sat, 27 Aug 2011 05:04:47 +0000 (05:04 +0000)]
Modify make.m so that it is workable
for MATLAB and OCTAVE under Windows, Mac, and Unix.
Modify "linear.h" into "../linear.h".
Handle the varible MX_API_VER of OCTAVE.
popo [Wed, 29 Jun 2011 13:34:11 +0000 (13:34 +0000)]
Modify linear.cpp so that users can easily give initial solutions.
popo [Sat, 25 Jun 2011 19:54:30 +0000 (19:54 +0000)]
Add soname in the shared library.
popo [Fri, 27 May 2011 03:37:10 +0000 (03:37 +0000)]
use "WARNING" and "ERROR" for warning messages and error messages respectively.
popo [Mon, 9 May 2011 23:37:09 +0000 (23:37 +0000)]
Add dependency for tron.o, linear.o, and blas.a in Makefile and matlab/Makefile.
popo [Sun, 10 Apr 2011 14:57:03 +0000 (14:57 +0000)]
Remove mxIsSparse(prhs[1]) in do_predict(...) in matlab/predict.c. Sparse format is checked in mexFunction(...).
cjlin [Sat, 26 Mar 2011 03:29:01 +0000 (03:29 +0000)]
copyright: 2010 -> 2011
eaudex [Sat, 26 Mar 2011 02:04:49 +0000 (02:04 +0000)]
Change libsvmread/write.mexw64 to symbolic links to
/home/faculty/cjlin/software/svm/libsvm/libsvm/trunk/windows/
libsvmread.mexw64 and libsvmwrite.mexw64
coha [Fri, 25 Mar 2011 09:55:12 +0000 (09:55 +0000)]
Build windows binaries, delete interface of 32 bits and add new matlab interface (only 64 bits)
coha [Fri, 25 Mar 2011 09:50:34 +0000 (09:50 +0000)]
Build windows binaries and add new matlab interface (only 64 bits)
popo [Wed, 16 Mar 2011 03:12:39 +0000 (03:12 +0000)]
matlab/README is modified because 32bit Windows binary files are going to be replaced by 64bit Windows binary files in next version.
eaudex [Mon, 14 Mar 2011 05:02:26 +0000 (05:02 +0000)]
@@ -1562,8 +1562,6 @@
In solve_l1_lr(.),
iter++;
- //if(iter % 10 == 0)
- // info(".");
if(QP_Gnorm1_new <= inner_eps*Gnorm1_init)
{
@@ -1573,7 +1571,6 @@
//active set reactivation
else
{
- info("*");
QP_active_size = active_size;
eaudex [Sat, 12 Mar 2011 17:28:47 +0000 (17:28 +0000)]
In solve_l1_lr(.),
change
cond = w_norm_new-w_norm - sigma*delta;
cond += negsum_xTd;
to
cond = w_norm_new - w_norm + negsum_xTd - sigma*delta;
popo [Fri, 11 Mar 2011 19:32:49 +0000 (19:32 +0000)]
help message in matlab/train.c is modified according to trunk/train.c.
popo [Wed, 9 Mar 2011 19:06:20 +0000 (19:06 +0000)]
The default print function for matlab interface should be mexPrintf(.).
eaudex [Mon, 7 Mar 2011 08:21:31 +0000 (08:21 +0000)]
In solve_l1r_lr(.),
replace CDN with newGLMNET.
In solve_l1r_l2_svc(.),
change the stopping condition from inf-norm to 1-norm.
cttsai [Sat, 5 Mar 2011 08:46:26 +0000 (08:46 +0000)]
- Rename linear.py and linearutil.py to liblinear.py and liblinearutil.py.
- Add L2R_LR_DUAL to python interface.
popo [Sat, 26 Feb 2011 00:09:48 +0000 (00:09 +0000)]
matlab/README is slightly modified. (Section 'Other Utilities' and 'Examples' are swapped.)
cjlin [Thu, 24 Feb 2011 04:41:40 +0000 (04:41 +0000)]
add libsvmwrite.mexw32 and libsvmread.mexw32
cjlin [Thu, 24 Feb 2011 04:38:14 +0000 (04:38 +0000)]
modify symlinkx for libsvmwrite and libsvmread
popo [Wed, 23 Feb 2011 22:57:08 +0000 (22:57 +0000)]
In matlab interface, tron.o and linear.o are in the parent directory now.
matlab/README is also slightly modified.
popo [Wed, 16 Feb 2011 13:35:05 +0000 (13:35 +0000)]
Change the path of loading *.dll and *.so to the relative path of linear.py.
popo [Mon, 14 Feb 2011 15:31:57 +0000 (15:31 +0000)]
In solve_l2r_lr_dual, the if-condition "newton_iter < l/10" is replaced by "newton_iter <= l/10." Otherwise, the inner epsilon may never be changed for l < 10.
xflash96 [Sat, 5 Feb 2011 12:21:49 +0000 (12:21 +0000)]
print error when 1. no label 2. emtpy line
cjlin [Mon, 11 Oct 2010 05:24:38 +0000 (05:24 +0000)]
add a comment in l1r_lr indicating that
approx line search condition works only
if xij>=0
popo [Mon, 13 Sep 2010 04:56:15 +0000 (04:56 +0000)]
windows binaries are uploaded.
popo [Mon, 13 Sep 2010 03:52:03 +0000 (03:52 +0000)]
Remove a duplicated ''options:'' in README.
popo [Fri, 10 Sep 2010 08:12:27 +0000 (08:12 +0000)]
check model_ptr != NULL before calling free_model_content in free_and_destroy_model
cjlin [Sun, 5 Sep 2010 13:57:28 +0000 (13:57 +0000)]
add LR dual for matlab interface
cjlin [Sun, 5 Sep 2010 13:38:46 +0000 (13:38 +0000)]
- for (s=0;s<active_size;s++)
+ for (s=0; s<active_size; s++)
cjlin [Sun, 5 Sep 2010 13:36:05 +0000 (13:36 +0000)]
Add L2-regularized logistic regression (dual)
by implementing Alg 5 of Yu et al.