This code is my implementation (in Python) of the methods presented in the paper:
A Sparse-group Lasso. Noah Simon, Jerome Friedman, Trevor Hastie, Rob Tibshirani http://www.stanford.edu/~hastie/Papers/SGLpaper.pdf
*_semisparse variants of the methods correspond to cases where one would allow sparsity for some dimensions but not all (for the L1-norm penalty).
In these cases, an indicator vector ind_sparse should be given that has 0 values for dimensions that should not be pushed towards sparsity and 1 values otherwise.
Typical function calls are given in the test_sgl.py script (models are sklearn-like objects).
For more information, you can also refer to this notebook.