Clear Filters
Clear Filters

fitrlinear for large data set

3 views (last 30 days)
Alessandro Fassò
Alessandro Fassò on 25 Feb 2021
Commented: Aditya Patil on 29 Mar 2021
I am trying a large regression/lasso model with n=90000 rows and p=500 columns
[mhat,FitInfo]=fitrlinear(X,y,'Learner','leastsquares');
I tryied also additional parameters
'solve','sparsa'
'Regularization','lasso'
The problem is that, when X has 200 columns or more, all the elements of mhat.Beta are ZERO
Do you have any suggestion about that?
Thanks,
Alessandro

Answers (1)

Aditya Patil
Aditya Patil on 29 Mar 2021
With high dimensional data, it is expected that some of the predictors won't have much effect on the response.
As a workaround, you can try to reduce the dimension using Dimensionality Reduction and Feature Extraction techniques.
  2 Comments
Alessandro Fassò
Alessandro Fassò on 29 Mar 2021
Thanks for your answer!
I agree that "some of the predictors won't have much effect ...", but I expect that others do have an effect (I know from preliminary correlation analysis and maller regression excercises).
Note that X has rank > 200.
The problem is that fitrlinear give me ALL the betas=0. It comes very fast despite the large dimension problem.
Of course one can perform some preliminary dimensionality reduction, but I expect this is made by the lasso option of fitrlinear, I tried in various exercises like
>> fitrlinear(..., 'regularization','lasso','lambda',lambda);
for various lambda.
Aditya Patil
Aditya Patil on 29 Mar 2021
Can you provide the data so that I can reproduce the issue? Also provide the output of the version command.

Sign in to comment.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!