add MultiTaskLassoCV and MultiTaskLasso

[agramfort]

@mathurinm I tried to add MultiTaskLassoCV and MultiTaskLasso to help @ja-che with his experiments but the test does not end. It seems like it's not converging.

@mathurinm can you have a look?

also you'll see that mtl_path is not consistent in API with sklearn lasso_path

wdyt?

[mathurinm]

I removed fit_intercept=True testing (not supported yet, do you need it for this project?)

I used the datafit scaling by n_samples in the cython code to harmonize with both celer's lasso and sklearn's MTL.

Now the tests pass locally.

By the way do you know why Travis is not triggered?

[mathurinm]

I will configure travis to trigger builds for fork's PR. It is ongoing maintenance so I will do it later. Thank you @agramfort

[agramfort]

works for me !

do you want to activate CIs before merging to avoid breaking something?

[mathurinm]

yes :)

[agramfort]

push an empty line and let's see?

…

[codecov-io]

Codecov Report

Merging #114 into master will increase coverage by 2.04%.
The diff coverage is 94.20%.

@@            Coverage Diff             @@
##           master     #114      +/-   ##
==========================================
+ Coverage   79.17%   81.22%   +2.04%     
==========================================
  Files          11       11              
  Lines         658      719      +61     
  Branches       97       98       +1     
==========================================
+ Hits          521      584      +63     
+ Misses        105      104       -1     
+ Partials       32       31       -1     

Impacted Files	Coverage Δ
celer/homotopy.py	81.29% <60.00%> (+3.35%)	⬆️
celer/dropin_sklearn.py	96.22% <92.85%> (-1.28%)	⬇️
celer/tests/test_lasso.py	100.00% <100.00%> (ø)
celer/tests/test_mtl.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update be79116...55b4e46. Read the comment docs.

[ja-che]

Thanks for the help @mathurinm. In our experiment, we only need fit_intercept=False.

[agramfort]

@ja-che I am trying your code with the new code but so far not clear how much it helps. I may need @mathurinm to look :(

…

[mathurinm]

Keep me posted about the numericq if I can give it a more detailed look