[scikit-learn] LassoCV.coef not implemented (I think)

[Guillaume Lemaître]

On Sun, 31 Jan 2021 at 21:24, Robert Slater <rdslater at gmail.com> wrote:

> Appreciate the clarification.  I definitely think the docs need some
> polish as coef_ only returns a single fitting of coefficients and not the
> coefficients along the path as stated in the api guide.
>

I am confused here. LassoCV states:

*coef : *ndarray of shape (n_features,) or (n_targets, n_features)

Parameter vector (w in the cost function formula).
So it seems exactly what it is returning. It does not return the
coefficients along the path.
Which documentation are you referring to when stating the API guide (if you
could provide a link, it would be really helpful)?


> I am seeing
>
> alpha_
> alphas_
> coef_
> dual_gap_
>
> as fitted variables (plus a few more) which is slightly different than the
> guide/api docs (all the names are plural in the api guide)
>
> I don't know if there is way to contribute an edit to the docs, I'd be
> more than happy to do it (Sorry I'm very OCD about such things, and I know
> this is a minor details)., I'd be happy to suggest the edit through proper
> channels.
>

You can always open a PR in the GitHub scikit-learn repository because the
documentation is actually the docstring from the classes and functions.
The user guide documentation is located in the /doc folder and the
contributing guide will be helpful to start with:
https://scikit-learn.org/stable/developers/contributing.html


>
> On Sun, Jan 31, 2021 at 2:02 PM Guillaume Lemaître <g.lemaitre58 at gmail.com>
> wrote:
>
>> Hi Robert,
>>
>> > I do have a .coef_ variable which I believe is the coefficient for the
>> best fit only.
>>
>> `coef` never existed. Fitted attributes always end with underscore.
>> We do not store coefficients for all fitted `alphas_`.
>> We provide some information regarding the MSE path for all tried alphas:
>> https://scikit-learn.org/stable/auto_examples/linear_model/plot_lasso_model_selection.html
>>
>> > the  alphas and alphas_ variables have a similar issue in that alphas
>> returns nothing while alphas_ returns the list of alphas used.
>>
>> You probably created a model such as `model = LasssoCV()`. By default,
>> the parameter `alpha=None` thus accessing it will return None. After
>> fitting,
>> `alphas_` will be automatically created as specified in the
>> documentation. It will correspond to the values tried by cross-validation.
>> If instead, you are passing an array to `alphas` then `alphas_` will be
>> the same as `alphas_` after calling `fit`.
>>
>> Cheers,
>>
>>
>> On Sun, 31 Jan 2021 at 20:45, Robert Slater <rdslater at gmail.com> wrote:
>>
>>> I was writing an example for my students when I came across what I think
>>> is an issue.  In version 24.1 using the LassoCV, the <object>.coef variable
>>> should have a list of my coeficeients (at least according to my
>>> understanding of the documents).  However, the variable is not populated
>>> nad throws an error
>>>
>>> 'LassoCV' object has no attribute 'coef'
>>>
>>>
>>> I do have a .coef_ variable which I believe is the coefficient for the
>>> best fit only.
>>>
>>> the  alphas and alphas_ variables have a similar issue in that alphas
>>> returns nothing while alphas_ returns the list of alphas used.
>>>
>>> I'm not sure if this is an documentation oversight or a real issue but
>>> wanted to get clarification.
>>>
>>> I can get what I need from o ther methods, but wanted to see if this
>>> needed to be addressed.
>>>
>>> Best Regards,
>>>
>>> Robert Slater
>>> _______________________________________________
>>> scikit-learn mailing list
>>> scikit-learn at python.org
>>> https://mail.python.org/mailman/listinfo/scikit-learn
>>>
>>
>>
>> --
>> Guillaume Lemaitre
>> Scikit-learn @ Inria Foundation
>> https://glemaitre.github.io/
>> _______________________________________________
>> scikit-learn mailing list
>> scikit-learn at python.org
>> https://mail.python.org/mailman/listinfo/scikit-learn
>>
> _______________________________________________
> scikit-learn mailing list
> scikit-learn at python.org
> https://mail.python.org/mailman/listinfo/scikit-learn
>


-- 
Guillaume Lemaitre
Scikit-learn @ Inria Foundation
https://glemaitre.github.io/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.python.org/pipermail/scikit-learn/attachments/20210131/cca21342/attachment.html>