* Woops, my previous reply got munged up, so I'm resubmitting it. Please
ignore my previous messed up email.
> 2012/10/30 Afik Cohen :
> >> Do you know what they are doing? I would expect they just do a soft-max.
> > I don't. :) But according to the LIBLINEAR FAQ: "If you really would like to
>
Hi Lars,
Thanks for your reply.
>
> 2012/10/30 Afik Cohen :
> >> Do you know what they are doing? I would expect they just do a soft-max.
> > I don't. :) But according to the LIBLINEAR FAQ: "If you really would like to
> > have probability outputs for SVM in LIBLINEAR, you can consider using th
2012/10/30 Afik Cohen :
>> Do you know what they are doing? I would expect they just do a soft-max.
> I don't. :) But according to the LIBLINEAR FAQ: "If you really would like to
> have probability outputs for SVM in LIBLINEAR, you can consider using the
> simple
> probability model of logistic re
Hello Andreas,
Thanks for the reply.
> Why do you want to use probability estimates in liblinear?
We need probability estimates because our use case requires a way of gauging how
'confident' the match is, when the classifier chooses a category for an input.
> Do you know what they are doing? I wo
2012/10/30 Afik Cohen :
> Now, however, we've run into a problem when we tried to upgrade to
> scikit-learn 0.13. It appears there have been significant changes to the
> underlying LIBLINEANR library as well as changes to the svm/classes
> interfaces;
> a recent commit shows almost 4000 lines
Hi Afik.
Thanks for your mail.
Why do you want to use probability estimates in liblinear?
Do you know what they are doing? I would expect they just do a soft-max.
This is not really "probability output", it is just a way to normalize
the decision
function.
This could be very easily implemented in
Hi all,
We've been using scikit-learn 0.12 to train LIBLINEAR's implementation of
LinearSVC. We require probability estimates for each prediction, and this isn't
supported out of the box by LinearSVC, so I emailed LIBLINEAR's author,
Dr. Chih-Jen Lin, for assistance. He showed us that LIBLIN
On Tue, Oct 30, 2012 at 08:55:04PM +, Andreas Mueller wrote:
> Does the PR now actually include the blas3 sources?
> It didn't have before, which was a sort of separate issue from the segfault.
Good point. Sorry for the confusion. So there are 2 issues was blas3:
avoiding segfaults, and includ
On 10/30/2012 01:43 PM, Gael Varoquaux wrote:
> On Tue, Oct 30, 2012 at 01:38:48PM +, Andreas Mueller wrote:
>> I tried working on the neural network a bit more and I think it would be
>> good to use
>> blas3 calls there. Unfortunately, these are not included in the blas
>> that sklearn ships.
2012/10/30 Gael Varoquaux :
> On Mon, Oct 29, 2012 at 05:05:11PM -0400, David Warde-Farley wrote:
>> Comment from the peanut gallery: this seems like a supremely odd
>> function to have, since int32s cannot be safely represented in 32-bit
>> floating point anyway (nor int64s in float64, but there's
On Tue, Oct 30, 2012 at 09:55:25AM -0400, Frédéric Bastien wrote:
> scipy wrap those call, but I don't know if this is an allowed
> dependency in scikit-learn.
It is. But we want them at the C level (cython level, actually).
G
-
Hi,
scipy wrap those call, but I don't know if this is an allowed
dependency in scikit-learn.
Fred
On Tue, Oct 30, 2012 at 9:43 AM, Gael Varoquaux
wrote:
> On Tue, Oct 30, 2012 at 01:38:48PM +, Andreas Mueller wrote:
>> I tried working on the neural network a bit more and I think it would b
On Tue, Oct 30, 2012 at 01:38:48PM +, Andreas Mueller wrote:
> I tried working on the neural network a bit more and I think it would be
> good to use
> blas3 calls there. Unfortunately, these are not included in the blas
> that sklearn ships.
> This problem already came up for the euclidean d
fabian is the right person to ask this.
Alex
On Tue, Oct 30, 2012 at 2:38 PM, Andreas Mueller
wrote:
> Hi everybody.
> I tried working on the neural network a bit more and I think it would be
> good to use
> blas3 calls there. Unfortunately, these are not included in the blas
> that sklearn ship
Hi everybody.
I tried working on the neural network a bit more and I think it would be
good to use
blas3 calls there. Unfortunately, these are not included in the blas
that sklearn ships.
This problem already came up for the euclidean distance speedup PR.
I'm not sure where to get the sources fr
On Mon, Oct 29, 2012 at 05:05:11PM -0400, David Warde-Farley wrote:
> Comment from the peanut gallery: this seems like a supremely odd
> function to have, since int32s cannot be safely represented in 32-bit
> floating point anyway (nor int64s in float64, but there's not much you
> can do about it w
Gael,
Thanks for let me know.
Didier
-Original Message-
From: Gael Varoquaux [mailto:gael.varoqu...@normalesup.org]
Sent: 30 October 2012 10:31
To: scikit-learn-general@lists.sourceforge.net
Subject: Re: [Scikit-learn-general] Dynamic Time Warping measure (DTW)
On Tue, Oct 30, 2012 at
On Tue, Oct 30, 2012 at 10:08:19AM -, Didier Vila wrote:
> So, I will be an active reader/tester/ user of the implementation of the DTW
> once everything is in scikit learn.
I am not sure that DTW is in the scope of scikit-learn right now: the
scikit still has no good way of dealing with time
Adrien,
Thanks to your email.
At the moment, I never worked in cython. (And I don't know how to use it).
So, I will be an active reader/tester/ user of the implementation of the DTW
once everything is in scikit learn.
All, Can someone let me know which GitHub Repertory I should follow for the
A very good and efficient generalization of DTW is the Global Alignment
Kernel of Marco Cuturi:
http://www.iip.ist.i.kyoto-u.ac.jp/member/cuturi/GA.html
I did a small cython wrapper if you want to try it out:
http://www.iip.ist.i.kyoto-u.ac.jp/member/cuturi/Code/TGA_python_wrapper_v1.0.tar.gz
C
20 matches
Mail list logo
