I've just sent a PR which implements Oliviers solution - The overhead
(converting to f-style) only applies to multi-class classification so
I think its ok to do it this way. SGDClassifier currently does not
support `partial_fit` only via the `init_coef` and `init_intercept`
arguments, they are
On Mon, Jan 9, 2012 at 2:00 AM, Olivier Grisel olivier.gri...@ensta.org wrote:
Hi all,
As discussed earlier, here is a new tool to publish the doc on github
rather than sourceforge.
The result is available here:
https://github.com/scikit-learn/scikit-learn.org (the repo for the
tool,
Thanks :)
Can someone switch the DNS?
--
Olivier
--
Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex
infrastructure or vast IT resources to deliver seamless, secure access to
virtual desktops.
On Mon, Jan 09, 2012 at 11:11:29AM +0100, Lars Buitinck wrote:
2012/1/9 Fabian Pedregosa fabian.pedreg...@inria.fr:
I'm OK with the idea of having one class for classification and one
for regression. It's conceptually easier and simplifies the docs. +1
What should the parameter be called
2012/1/9 Gael Varoquaux gael.varoqu...@normalesup.org:
On Mon, Jan 09, 2012 at 11:11:29AM +0100, Lars Buitinck wrote:
2012/1/9 Fabian Pedregosa fabian.pedreg...@inria.fr:
I'm OK with the idea of having one class for classification and one
for regression. It's conceptually easier and
2012/1/9 Fabian Pedregosa fabian.pedreg...@inria.fr:
On Mon, Jan 9, 2012 at 2:00 AM, Olivier Grisel olivier.gri...@ensta.org
wrote:
Hi all,
As discussed earlier, here is a new tool to publish the doc on github
rather than sourceforge.
The result is available here:
On Mon, Jan 9, 2012 at 7:40 PM, Gael Varoquaux
gael.varoqu...@normalesup.org wrote:
Can it be that if nu is None C is used, or do you thihnk that this is
confusing?
That wouldn't work for NuSVR, which uses both C and nu (I know, it's confusing).
If we merge C-SVM and nu-SVM classes, users may
On Mon, Jan 09, 2012 at 08:19:33PM +0900, Mathieu Blondel wrote:
Can it be that if nu is None C is used, or do you thihnk that this is
confusing?
That wouldn't work for NuSVR, which uses both C and nu (I know, it's
confusing).
That's exactly what I had in mind: I was proposing nu to be the
2012/1/9 Olivier Grisel olivier.gri...@ensta.org:
That would work. Alternatively we could have:
algorithm=csvm or algorithm=nusvm as already used elsewhere
(e.g. neighbors, pls, dict_learning and manifold).
+1.
--
Lars Buitinck
Scientific programmer, ILPS
University of Amsterdam
Is decomposition.sparse_encode implemented using SPAMS (
http://www.di.ens.fr/~mairal/software_eng.php ) ? If not, does anyone
know how the two implementations compare?
Thanks,
Ian
--
Ridiculously easy VDI. With Citrix
Short answer, no.
sparse_encode is just a wrapper for funcionality that existed in the
scikit already (lasso, omp), with support for parallelization. We
couldn't embed SPAMS anyway, because of the license IIRC.
A benchmark would be interesting indeed.
Vlad
On 09.01.2012, at 18:02, Ian
2012/1/9 Mathias Verbeke mathi...@gmail.com:
Dear all,
In the documentation of the SVM module, I saw that it was possible to pass
your own Gram matrix to the kernel. I was wondering if it was also possible
to do the reverse, i.e. to export the calculated Gram matrix (that gives the
what is implemented in the scikit matches the algorithms implemented in SPAMS.
SPAMS implements more like positivity constraints for example.
A real comparison in terms of speed and results has however not been done.
Note that for the positivity constraint we just need to modify
LassoLARS and
13 matches
Mail list logo