Initial Active Set¶
User-specified initial active set¶
We believe that it worth allowing given an initial active set so that the splicing process starts from this set for each sparsity. It might come from prior analysis, whose result is not quite precise but better than random selection, so the algorithm can run more efficiently. Or you just want to give different initial sets to test the stability of the algorithm.
Note that this is NOT equivalent to
always_select, since they can be exchanged to inactive set when splicing.
To specify initial active set, an additive argument
A_init should be
LinearRegression(A_init=[0, 1, 2], support_size=range(0, 5))In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
LinearRegression(A_init=[0, 1, 2], support_size=range(0, 5))
Some strategies for initial active set are:
sparsity = len(A_init), the splicing process would start from
sparsity > len(A_init), the initial set includes
A_initand other variables with larger forward sacrifices chooses.
sparsity < len(A_init), the initial set includes part of
A_initwill only affect splicing under the first sparsity in
A_initwill affect each fold but not the re-fitting on full data.
abess R package also supports user-defined initial active set.
For R tutorial, please view
Total running time of the script: (0 minutes 0.003 seconds)