[MRG+1] Ensure coef_ is an ndarray when fitting LassoLars #8160
Conversation
|
Great, it looks like you've got most of the way to a regression test at #1615. Please add it? |
|
Tests added. |
|
(I've confirmed the test fails in master) |
| @@ -689,6 +689,8 @@ def fit(self, X, y, Xy=None): | |||
| a[0] for a in (self.alphas_, self.active_, self.coef_path_, | |||
| self.coef_)] | |||
| self.n_iter_ = self.n_iter_[0] | |||
| else: | |||
| self.coef_ = np.array(self.coef_) | |||
There was a problem hiding this comment.
it would be more efficient to preallocate self.coef_ to an empty array rather than first having a list than copying the data to an array
There was a problem hiding this comment.
True. Would you like me to add that optimization to this PR?
There was a problem hiding this comment.
If you can do it, it will be great :)
|
LGTM see why CIs complain but diff looks clean |
|
@agramfort Error on circleci are not related. We have the same problem on several PR (ex : #8135). |
|
@perimosocordiae Ci problems seems solved now, can you rebase ? |
|
Nah. Merging. |
|
@jnothman Thanks :) |
|
@perimosocordiae Thanks for your work |
) * Fix scikit-learngh-1615: ensure self.coef_ is an ndarray
) * Fix scikit-learngh-1615: ensure self.coef_ is an ndarray
) * Fix scikit-learngh-1615: ensure self.coef_ is an ndarray
) * Fix scikit-learngh-1615: ensure self.coef_ is an ndarray
) * Fix scikit-learngh-1615: ensure self.coef_ is an ndarray
This PR fixes #1615, an issue related to fitting a
LassoLarsmodel with the specific combination offit_path=True,fit_intercept=False, and >1 targets.The bug was masked in the
fit_intercept=Truecase because of this line,in which the coefficient list is promoted to an ndarray by division with another ndarray.