Previously plot_sparse_recovery.py had 10k lines output. I don't think you want to know about some of your randomized designs being ill-conditioned.
Also: this was a pain. the class-structure might need some cleaning up ;)
What is wrong with using the warnings module to do this (and recommending same to users in the documentation):
We could, but I find it less pretty. This is also not pretty in the code, but nicer for the user.
I think for the randomized version catching should be the default and I working with warnings is often brittle.
Somehow I wouldn't like to have code in RandomizedLasso that modifies the warning stack.
LGTM although you need to fix doctests to make travis happy
Hum, so depends on if we go with catching the warnings or adding the option.
Ok, but would you then also catch them by default in RandomizedLasso or leave them to the user?
Catch ConvergenceWarning in RandomizedL1
👍 for merge. Thanks
Coverage increased (+0.0%) when pulling d146d23 on amueller:less_loud_randomized_lasso into be5826d on scikit-learn:master.
LGTM as well. Merging.