Re: [Scikit-learn-general] scikit-learn wrapper in NLTK

2012-01-14 Thread Gael Varoquaux
On Sat, Jan 14, 2012 at 08:18:07PM +0100, Lars Buitinck wrote: > I already mentioned this in an email earlier today, but in case not > everyone read that: the Python Natural Language Toolkit (NLTK) now has > a wrapper for scikit-learn classifiers in its bleeding-edge version. That's really, really

Re: [Scikit-learn-general] scikit-learn wrapper in NLTK

2012-01-14 Thread Nelle Varoquaux
On 14 January 2012 20:23, Olivier Grisel wrote: > 2012/1/14 Lars Buitinck : > > Dear all, > > > > I already mentioned this in an email earlier today, but in case not > > everyone read that: the Python Natural Language Toolkit (NLTK) now has > > a wrapper for scikit-learn classifiers in its bleedi

Re: [Scikit-learn-general] scikit-learn wrapper in NLTK

2012-01-14 Thread Olivier Grisel
2012/1/14 Lars Buitinck : > Dear all, > > I already mentioned this in an email earlier today, but in case not > everyone read that: the Python Natural Language Toolkit (NLTK) now has > a wrapper for scikit-learn classifiers in its bleeding-edge version. I > hope this will make scikit-learn easier t

[Scikit-learn-general] scikit-learn wrapper in NLTK

2012-01-14 Thread Lars Buitinck
Dear all, I already mentioned this in an email earlier today, but in case not everyone read that: the Python Natural Language Toolkit (NLTK) now has a wrapper for scikit-learn classifiers in its bleeding-edge version. I hope this will make scikit-learn easier to use for NLP people; it mainly tran

Re: [Scikit-learn-general] One class svm change result if dataset is shuffled

2012-01-14 Thread Olivier Grisel
2012/1/14 Gael Varoquaux : > On Sat, Jan 14, 2012 at 01:40:53PM +0100, Olivier Grisel wrote: >> Do you think that the default tolerance set in scikit-learn is inadequate? > > No. Datasets for which it doesn't work well are, in my experience, > aberrant datasets for which the optimization problem is

Re: [Scikit-learn-general] One class svm change result if dataset is shuffled

2012-01-14 Thread Gael Varoquaux
On Sat, Jan 14, 2012 at 01:40:53PM +0100, Olivier Grisel wrote: > Do you think that the default tolerance set in scikit-learn is inadequate? No. Datasets for which it doesn't work well are, in my experience, aberrant datasets for which the optimization problem is ill-posed. Gael

Re: [Scikit-learn-general] One class svm change result if dataset is shuffled

2012-01-14 Thread Olivier Grisel
2012/1/14 Gael Varoquaux : > On Wed, Dec 21, 2011 at 06:36:03PM -, luca.fias...@iwr.uni-heidelberg.de > wrote: >> first thing I would like to say that I'm not so experienced with python >> therefore I might do something really stupid which I cannot see. >> Nevertheless I don't manege to unders

Re: [Scikit-learn-general] Adaptive Lasso

2012-01-14 Thread Olivier Grisel
Very interesting. I added an issue to track this: https://github.com/scikit-learn/scikit-learn/issues/555 -- RSA(R) Conference 2012 Mar 27 - Feb 2 Save $400 by Jan. 27 Register now! http://p.sf.net/sfu/rsa-sfdev2dev2 _

Re: [Scikit-learn-general] Sparse feature matrix generation

2012-01-14 Thread Lars Buitinck
2012/1/6 Dhruvkaran Mehta : > Is there a convenient way in scikits to go from "string" features like: ... > to a numpy matrix like: ... At present, no. But I donated a scikit-learn wrapper to NLTK yesterday, which translates NLTK's featuresets (dicts mapping feature names to numeric or boolean

Re: [Scikit-learn-general] Adaptive Lasso

2012-01-14 Thread Stéfan van der Walt
On Sat, Jan 14, 2012 at 2:38 AM, Alexandre Gramfort wrote: > I've put up a demo of adaptive lasso a.k.a. reweighed l1: > > https://gist.github.com/1610922 Cute! Do you by any chance have a reference that gives an intuitive motivation for the calculation of the weights? Stéfan -

Re: [Scikit-learn-general] One class svm change result if dataset is shuffled

2012-01-14 Thread Gael Varoquaux
On Wed, Dec 21, 2011 at 06:36:03PM -, luca.fias...@iwr.uni-heidelberg.de wrote: > first thing I would like to say that I'm not so experienced with python > therefore I might do something really stupid which I cannot see. > Nevertheless I don't manege to understand why the following script thro

[Scikit-learn-general] Adaptive Lasso

2012-01-14 Thread Alexandre Gramfort
hi sklearners, I've put up a demo of adaptive lasso a.k.a. reweighed l1: https://gist.github.com/1610922 If anyone is willing to make it a proper estimator in the main code base please do so. Cheers, Alex -- RSA(R) Con

Re: [Scikit-learn-general] One class svm change result if dataset is shuffled

2012-01-14 Thread Alexandre Gramfort
hi, it works if you change the tolerance of the optimality check. Set tol=1e-9: >>> clf = svm.OneClassSVM(nu=0.1, kernel="rbf", gamma=0.1, tol=1e-9) Alex PS : next time use a gist on github to avoid pasting code in an email. On Wed, Dec 21, 2011 at 7:36 PM, wrote: > Hi All, > first thing I

[Scikit-learn-general] Sparse feature matrix generation

2012-01-14 Thread Dhruvkaran Mehta
Hi scikit-learn users, *Is there a convenient way in scikits to go from "string" features like:* "uc_berkeley", "google", 1 "stanford", "intel", 1 . . . "uiuc", "texas_instruments", 0 *to a numpy matrix like:* "uc_berkeley", "stanford", ..., "uiuc", "google", "intel", "texas_instruments", "boo

[Scikit-learn-general] One class svm change result if dataset is shuffled

2012-01-14 Thread luca.fiaschi
Hi All, first thing I would like to say that I'm not so experienced with python therefore I might do something really stupid which I cannot see. Nevertheless I don't manege to understand why the following script throw an assertion error. Indeed, training the one class svm with a dataset or with a s