WIP: OrthogonalMatchingPursuitCV #1042
Conversation
|
Cool! Thanks for tackling this Vlad! |
| @@ -145,6 +164,10 @@ def _gram_omp(Gram, Xy, n_nonzero_coefs, tol_0=None, tol=None, | |||
| Indices of the positions of the elements in gamma within the solution | |||
| vector | |||
|
|
|||
| coefs, array, shape = (n_features, n_nonzero_coefs) | |||
| The first k values of column k correspond to the coefficient value | |||
| for the active features at that step. | |||
There was a problem hiding this comment.
can you add that it is returned only if return_path=True?
|
almost there :) |
|
Hi @vene, what is the state of this pr? |
|
@arjoly : It seems I forgot about it halfway, and I don't remember why, but I don't think there are any obstacles. I cannot tackle it in the next couple of days, though, but if you don't finish it in the mean while, I'll do it then. |
|
Thanks! |
|
I'm working on this. |
| coef[idx[:n_active + 1], k, n_active] = x[:n_active + 1] | ||
| else: | ||
| x, idx = out | ||
| coef[idx, k] = x | ||
| return np.squeeze(coef) |
There was a problem hiding this comment.
@agramfort :
Here instead of alphas we can return np.arange(n_features), followed by active and coef and this should be usable as-is inside the _lars_path_residues function. Unfortunately there is duplication of normalization code and logic and OMP normalization is non-trivial when precomputing is in play. However I wonder whether many people actually use precomputing; and if so; in another context than sparse coding (where we already have a convenience function). Maybe we can simplify the logic a bit by disallowing some things, and allow power users to do it through functions instead of estimators.
|
LGTM to far. More when #2186 is merged. Tomorrow hopefully. |
|
A class which implements fit / predict and takes a cv object would be nice too (possibly in another PR). |
|
Mathieu, We're waiting for a cleaner infrastructure for path classes and generic cv On Tue, Jul 23, 2013 at 3:08 AM, Mathieu Blondel
|
DOC/FIX affinity_propagation damping default value.
DOC Clarify docstrings
FIX: work with max_iter=1, though it does not make sense DOC add OrthogonalMatchingPursuitCV docstrings
|
@agramfort @mblondel @arjoly I updated this. |
|
Couldn't get rid of some pesky commits without opening a new PR. #2224 |
As discussed in issue #1930
Todo: