Hi Fabien,
welcome to the list!
If you are interested in the exact locations of the kinks in the
coefficient path, you may prefer using e.g. LassoLarsCV. It works on your
size of problem (iff "highly collinear" doesn't mean "basically equal") and
has the attributed "coef_path_" (
https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/linear_model/least_angle.py#L1101
).
Highly correlated designs generally result in a strong support
identification instability of these algorithms. It depends on what you care
about to be able to conclude if the warnings you see are relevant or not.
If it is prediction and your variables are highly correlated, then
whichever is selected, they will predict similarly.
Hope this helps, although I'm not sure I am giving a constructive answer to
your question.
Michael
On Mon, Oct 6, 2014 at 10:05 AM, Fabien <[email protected]> wrote:
> Folks,
>
> this is my first message on this newsgroup, so first: Hi!
>
> I have two questions, I hope they are not too trivial:
>
> 1. Access to coefficients in LassoCV
> I use LassoCV to find the optimal alpha for my problem. For analysis
> purposes I'd like to get access to the paths coefficients, more or less
> like it's done in this old example here:
>
> http://scikit-learn.org/0.8/auto_examples/linear_model/plot_lasso_path_crossval.html
>
> I see that the coef_path_ attribute has been removed from lassoCV.
> What's the rationale behind this choice? To get the coefficients, should
> I use lasso_path and find the best MSE by myself, or did I miss
> something obvious here?
>
> 2. Convergence warnings
> My use case with Lasso is a "small n (48) large p (~100)" problem with
> some predictors highly collinear. I constantly get the following warning
> message when using LassoCV:
> "ConvergenceWarning: Objective did not converge. You might want to
> increase the number of iteration"
> Since the results look fine and LassoCV was able to find a MSE minimun,
> I guess that one (or more) of the models along the path had trouble to
> converge. I tried increasing iterations and tolerance thresholds without
> success. What should I do to solve this problem without awkward
> try-and-error steps? Which particular feature in my data could cause
> this warning?
>
> Thanks a lot!
>
> Fabien
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> ------------------------------------------------------------------------------
> Slashdot TV. Videos for Nerds. Stuff that Matters.
>
> http://pubads.g.doubleclick.net/gampad/clk?id=160591471&iu=/4140/ostg.clktrk
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>
------------------------------------------------------------------------------
Slashdot TV. Videos for Nerds. Stuff that Matters.
http://pubads.g.doubleclick.net/gampad/clk?id=160591471&iu=/4140/ostg.clktrk
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
