Hey Gael and Alex,
Thanks for getting back to me:
On Wed, Sep 26, 2012 at 12:42 AM, Alexandre Gramfort <
[email protected]> wrote:
> hi ariel,
>
> indeed coordinate descent (an all interative solvers I know) will
> converge slowly for low regularization. So just increase max_iter and
> set tol to 1e-15
I haven't tried this yet - I'll try it tomorrow. In a way it sounds like
it's inadvertently implementing an early stopping criterion, which is also
a form of regularization. That's confusing, considering that I set the
regularization to 0. But at least I think I might understand it now. I'll
try to play with the tolerance and I will report back.
> Best,
> Alex
>
> On Wed, Sep 26, 2012 at 7:36 AM, Gael Varoquaux
> <[email protected]> wrote:
> > Hi Ariel,
> >
> > On Tue, Sep 25, 2012 at 05:44:21PM -0700, Ariel Rokem wrote:
> >> Initially, I suspected that this has to do with the non-negativity
> >> constraint I applied, so I removed that.
> >
> > Indeed, if you are imposing positivity, you do not have a least square.
> >
>
I forgot to mention that I did also remove the non-negativity constraint,
but still didn't get it to fit all the way.
> >> Then, I was wondering whether it might have to do with the correlation
> >> among my regressors, so I tried linear_model.LinearRegression. This
> >> fits all the way, leaving no residuals.
> >
> > With no regularization, the algorithm used by the ElasticNet object (a
> > coordinate descent) has a very hard time converging. Are you sure that
> > the residuals that you are seeing are not simply numerical errors?
> >
>
They're not negligibly small, so I don't think you can call them numerical
error.
More tomorrow. Thanks in the meanwhile.
Cheers,
Ariel
------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://ad.doubleclick.net/clk;258768047;13503038;j?
http://info.appdynamics.com/FreeJavaPerformanceDownload.html
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
