any ideas why it not converge?
from http://www.csie.ntu.edu.tw/~cjlin/libsvm/faq.html, the author said
replacing the Qfloat float with double can help,
but it didn't work for me.
On Aug 7, 2013, at 9:09 PM, Olivier Grisel wrote:
> 2013/8/7 Gael Varoquaux :
>> On Wed, Aug 07, 2013 at 03:27:38PM
hi
Lasso and LassoLars support matrices. If you have too many zeros
use a smaller alpha or better use LassoCV or LassoLarsCV to set the
alpha by cross-validation. The LassoCV and LassoLarsCV do
not support matrices though. So use a for loop over the columns of Y.
HTH
Alex
On Wed, Aug 7, 2013 at
Great!
Thank you all for the hard work. The changelog is really impressive :)
Gilles
On 8 August 2013 01:18, Gael Varoquaux wrote:
> Hi Scikiters,
>
> I have tagged and pushed release 0.14.
>
> Vlad will build the Windows binaries tomorrow. In the mean time, the 0.14
> tarball is on pypi, but
Hi Scikiters,
I have tagged and pushed release 0.14.
Vlad will build the Windows binaries tomorrow. In the mean time, the 0.14
tarball is on pypi, but I have not hidden the old files for the 0.13.
The new website is live! Congratulations everybody who worked on it.
The master branch and the de
Hi
I would like to solve the standard Lasso optimization:
(1 / (2 * n_samples)) * ||Y - XW||^2_2 + alpha * ||W||_1
Only difference being that I want to take Y to be a matrix. This would
mean that W is also a matrix. Is it a good idea to use the lasso module
written in scikit-lea
On Wed, Aug 07, 2013 at 04:20:17PM +0200, Gael Varoquaux wrote:
> On Wed, Aug 07, 2013 at 04:18:14PM +0200, Olivier Grisel wrote:
> > Maybe we can release 0.14 with windows binaries for Python 2 only and
> > fix the Windows + Python 3 issues for a 0.14.1 release?
> I am still trying to work around
On Wed, Aug 07, 2013 at 04:18:14PM +0200, Olivier Grisel wrote:
> Maybe we can release 0.14 with windows binaries for Python 2 only and
> fix the Windows + Python 3 issues for a 0.14.1 release?
I am still trying to work around the problem, but indeed...
> I would need to download a whole bunch of
2013/8/7 Gael Varoquaux :
> On Wed, Aug 07, 2013 at 03:44:37PM +0200, Olivier Grisel wrote:
>> https://dl.dropboxusercontent.com/u/5743203/sklearn/scikit-learn-0.14a1.win32-py2.7.exe
>> (still currently uploading on my 3G tethered internet connection from
>> an island in Britany).
>
> Hum, but the
On Wed, Aug 07, 2013 at 03:44:37PM +0200, Olivier Grisel wrote:
> https://dl.dropboxusercontent.com/u/5743203/sklearn/scikit-learn-0.14a1.win32-py2.7.exe
> (still currently uploading on my 3G tethered internet connection from
> an island in Britany).
Hum, but the problem was with Python 3.3 (which
I just built a windows installer from the current 0.14.X branch with
windows c++ 2008 express on python 2.7 32bit:
https://dl.dropboxusercontent.com/u/5743203/sklearn/scikit-learn-0.14a1.win32-py2.7.exe
(still currently uploading on my 3G tethered internet connection from
an island in Britany).
P
2013/8/7 Gael Varoquaux :
> On Wed, Aug 07, 2013 at 03:27:38PM +0800, Terry Peng wrote:
>> Thanks, but i was using linear kernel.
>
> I don't really think that using a linear kernel with a OneClassSVM makes
> much sens. Do you have a specific intuition or reason to use a linear
> kernel?
I think i
Hey Alex.
So if you are using the ML for hiring, I think the least you could do is
write a testimonial for us ;)
They will soon be featured together with the logo on our frontpage:
https://github.com/scikit-learn/scikit-learn/pull/2323#issuecomment-22214484
(actually hey should be up as soon as
I tried yesterday evening to build the current master under Python 3.3 in
Windows using Anaconda, but I would get a segfault during one of the tests.
I didn't have too much time to investigate, so I suspect it was probably
because I hadn't configured my build environment correctly.
I'll try to se
Dear all,
In the sklearn.metrics module, the precision_recall_fscore_support and
precision_score have different default value for the "average" parameter, which
seems a bit counterintuitive -- but I did not go deep into the code. Is this
really the intended behavior ?
def precision_score(y_t
>
>b. A logo of INRIA (in a sliding carousel is good).
Just FYI, I'm doing this one quick
2013/8/6 Andreas Mueller
> Hi Gael.
> Sorry for the absence.
> I had to correct some exams for the last ~10h.
> I don't have a windows build environment, so I'm not sure I can be much
> help there. T
On Wed, Aug 07, 2013 at 03:27:38PM +0800, Terry Peng wrote:
> Thanks, but i was using linear kernel.
I don't really think that using a linear kernel with a OneClassSVM makes
much sens. Do you have a specific intuition or reason to use a linear
kernel?
G
--
Thanks, but i was using linear kernel. However the problem seems due to the
OneClassSVM not converged.
i.e. the obj = inf and rho is a very very big number if i set the verbose to
True and limit the number of the iteration.
But from the http://www.csie.ntu.edu.tw/~cjlin/libsvm/faq.html, the lib
17 matches
Mail list logo
