A few notes:
* Online and parallelisation are different things, both interesting and to
keep in mind, but that should not be confused.
* For non-negative matrix factorization, Julien Mairal's algorithm for
online dictionary learning can also be used (see the JMLR paper). It
needs a small mo
On Tue, Mar 26, 2013 at 1:41 AM, Olivier Grisel
wrote:
> I am also +1 a simple short term solution while still keeping longer
> terms goal for
> - proper multinomial penalized LR on one hand,
It would still be nice to have it but if Paolo is correct, we should
not expect it
to produce well-calib
On Wed, Mar 06, 2013 at 01:26:42PM -0500, Ronnie Ghose wrote:
> we can just put a javascript redirect / html redirect in the actual
> source
> code ...i'm pretty sure we can use some sort of engine or script to do this
> for
> us :/
That's very ugly, breaks search engines, and forces 2 HTTP
2013/3/25 Paolo Losi :
> My 2 cents ...
>
> The problem is that penalized (l1 o l2) single class Logistic Regression is
> not
> well calibrated to start with.
> In other terms: the penalization param value that optimizes classification
> accuracy is not
> guaranteed to be the one that maximizes pro
On Wed, Mar 06, 2013 at 10:25:26AM -0500, Satrajit Ghosh wrote:
> i don't know if this would be better a separate thread, but seemed related. is
> there a reason why we don't just move the docs to github?
Lack of .htaccess that prevent us from using the rewriterules that we use
to support multipl
My 2 cents ...
The problem is that penalized (l1 o l2) single class Logistic Regression is
not
well calibrated to start with.
In other terms: the penalization param value that optimizes classification
accuracy is not
guaranteed to be the one that maximizes probability estimation accuracy.
Optimiz
On Mon, Mar 25, 2013 at 10:57 PM, Lars Buitinck wrote:
> You mean the score rescaling at the end of section 2? What Zadrozny
> and Elkan really seem to be recommended is isotonic regression...
The question is how to deal with the *multiclass* case, which isotonic
regression cannot handle directl
On Wed, Mar 20, 2013 at 09:33:06AM +0100, [email protected] wrote:
> I need the random matrix to evaluate the predictiveness of a particular
> model - this time, not in terms of a "domain of applicability" :)
I would use the DummyClassifier.
G
--
2013/3/25 Mathieu Blondel :
>> I'm confused. Since when is that so? The other losses definitely support
>> OvR multi-class. I would be quite surprised if 'log' does not.
>
> predict_proba currently raises an exception in the multiclass case:
> https://github.com/scikit-learn/scikit-learn/blob/maste
Hello,
Thanks for the interest in taking part to the GSOC! :)
If they have time, I think that Gilles Louppe or Arnaud Joly would
make good mentors since they are from
University of Liège too. Gilles wrote his master thesis on recommender
systems so he has good knowledge of
matrix factorization pr
On Mon, Mar 25, 2013 at 9:19 PM, Andreas Mueller
wrote:
> I'm confused. Since when is that so? The other losses definitely support
> OvR multi-class. I would be quite surprised if 'log' does not.
predict_proba currently raises an exception in the multiclass case:
https://github.com/scikit-learn/
On 03/23/2013 06:37 PM, ShNaYkHs ShNaYkHs wrote:
> Well, I need a classifier in that can support partial_fit() and can
> give a probability for prediction proba_predict() (in case of
> multi-class, not only binary). PassiveAggressiveClassifier and
> Perceptron have no proba_predict() method, and
12 matches
Mail list logo
