Bug Fix: GLMCV does not pass group argument

[BeibinLi]

1. As above
2. Suppress warning for numpy type conversion.
3. TODO: need to update group lasso example, because it uses the "group" parameter.

[codecov-io]

Codecov Report

Merging #290 into master will not change coverage. The diff coverage is 100%.

@@           Coverage Diff           @@
##           master     #290   +/-   ##
=======================================
  Coverage   74.27%   74.27%           
=======================================
  Files           4        4           
  Lines         622      622           
  Branches      127      127           
=======================================
  Hits          462      462           
  Misses        123      123           
  Partials       37       37

Impacted Files	Coverage Δ
pyglmnet/pyglmnet.py	80.44% <100%> (ø)	:arrow_up:

---

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 7f1fbb7...d7c5594. Read the comment docs.

[BeibinLi]

The test on plot_group_lasso.py in CircleCI failed because it exceeds time limit (i.e. 10 minutes). The new group argument makes the training process longer. The GLM._prox(...) function and optimization will take more time than before.

We can solve this problem either by extending the time limit or adding print(...) during the training.

[jasmainak]

The test on plot_group_lasso.py in CircleCI failed because it exceeds time limit (i.e. 10 minutes). The new group argument makes the training process longer. The GLM._prox(...) function and optimization will take more time than before.

Does it converge / finish on your box though? Perhaps we could change the learning rate to make it converge faster?

[BeibinLi]

The test on plot_group_lasso.py in CircleCI failed because it exceeds time limit (i.e. 10 minutes). The new group argument makes the training process longer. The GLM._prox(...) function and optimization will take more time than before.

Does it converge / finish on your box though? Perhaps we could change the learning rate to make it converge faster?

Yes, it finished in my computer, which takes about one hour. (I did not count the time, but the final plot is nice and should be correct).

[jasmainak]

@BeibinLi can you update whats_new.rst to document your contribution?

[jasmainak]

Also, since the plot takes so long to be produced, can you rename the plot_grouplasso.py example and remove the plot prefix so that it is not built during the continuous integration? That way, we can at least catch other failing examples while we work on making this faster.

[jasmainak]

Merging so we can move on here. Thanks @BeibinLi !