Re: [Scikit-learn-general] building one global vector but used for all models

2012-10-04 Thread Joseph Turian
It appears that you aren't using the same vocabulary in the training and testing. Reading the docs, it seems you can trying using the "vocabulary" field in CountVectorizer, and having one vocabulary for all models. On Thu, Oct 4, 2012 at 1:39 AM, David Montgomery wrote: > Hi, > > I have this iss

Re: [Scikit-learn-general] libsvm PR

2012-10-04 Thread Joseph Turian
Well naturally, the most hilarious solution to this is to fork into a project called libsvm2. On Thu, Oct 4, 2012 at 4:48 PM, James Bergstra wrote: > Nothing so far. > > On Thu, Oct 4, 2012 at 4:15 PM, Joseph Turian wrote: >> What happened when you contacted the libsvm people? >> >> On Thu, Oct

Re: [Scikit-learn-general] Compressed sensing and Lasso: L1 penalization == L1 minimization?

2012-10-04 Thread Vlad Niculae
Hi Jaidev, This seems relevant to your question: http://metaoptimize.com/qa/questions/7897/are-lasso-and-basis-pursuit-really-the-same-thing Vlad On Oct 5, 2012, at 01:16 , Jaidev Deshpande wrote: > Hi, > > I've been going through the tomography reconstruction example > (http://scikit-learn.

Re: [Scikit-learn-general] Lasso path computations becomes unstable when X is a fat matrix

2012-10-04 Thread Alejandro Weinstein
On Thu, Oct 4, 2012 at 8:40 AM, Gael Varoquaux wrote: > On Thu, Oct 04, 2012 at 06:39:00AM -0600, Alejandro Weinstein wrote: >> Any reason for this behavior? > > It's probably because the lars algorithm is unstable with correlated > regressors. I tried with a home-made "naive" implementation of l

Re: [Scikit-learn-general] Lasso path computations becomes unstable when X is a fat matrix

2012-10-04 Thread Nelle Varoquaux
On 4 Oct 2012 16:41, "Gael Varoquaux" wrote: > > On Thu, Oct 04, 2012 at 06:39:00AM -0600, Alejandro Weinstein wrote: > > Any reason for this behavior? > > It's probably because the lars algorithm is unstable with correlated > regressors. This is a fundamental problem and is not really fixable. In

Re: [Scikit-learn-general] libsvm PR

2012-10-04 Thread James Bergstra
Nothing so far. On Thu, Oct 4, 2012 at 4:15 PM, Joseph Turian wrote: > What happened when you contacted the libsvm people? > > On Thu, Oct 4, 2012 at 3:41 PM, James Bergstra > wrote: >> On Sun, Sep 30, 2012 at 8:28 AM, Andreas Mueller >> wrote: >>> >>> Why do you want to rewrite the predi

Re: [Scikit-learn-general] libsvm PR

2012-10-04 Thread Joseph Turian
What happened when you contacted the libsvm people? On Thu, Oct 4, 2012 at 3:41 PM, James Bergstra wrote: > On Sun, Sep 30, 2012 at 8:28 AM, Andreas Mueller > wrote: >> >> >>> Why do you want to rewrite the predict code, which seems to be already >>> working? >>> (Doesn't this further divergenc

Re: [Scikit-learn-general] libsvm PR

2012-10-04 Thread James Bergstra
On Sun, Sep 30, 2012 at 8:28 AM, Andreas Mueller wrote: > > >> Why do you want to rewrite the predict code, which seems to be already >> working? >> (Doesn't this further divergence from the libsvm code base just >> increase the sklearn maintenance burden?) >> >> The key thing seems to be how hea

Re: [Scikit-learn-general] Lasso path computations becomes unstable when X is a fat matrix

2012-10-04 Thread Gael Varoquaux
On Thu, Oct 04, 2012 at 06:39:00AM -0600, Alejandro Weinstein wrote: > Any reason for this behavior? It's probably because the lars algorithm is unstable with correlated regressors. This is a fundamental problem and is not really fixable. In such a situation, use the 'Lasso' object, that relies on

[Scikit-learn-general] scikit[s]-image project rename

2012-10-04 Thread Stéfan van der Walt
Hey, folks After a discussion initiated by Mathieu Blondel, we decided to rename "scikits-image" to "scikit-image" in order to collectively present a more unified front for (kick-ass) scientific packages in Python. Check out our new website: http://skimage.org (http://scikit-image will also wor