Implementation of the K-support norm proximity operator #71
Conversation
Pull Request Test Coverage Report for Build 216
💛 - Coveralls |
|
Still Working on some tests to improve the coverage, but it would be cool to have a preliminary review |
| self.ridge = proximity.Ridge(linear.Identity(), weights) | ||
| self.elasticnet_alpha_0 = proximity.ElasticNet(linear.Identity(), | ||
| alpha=0, | ||
| beta=weights) | ||
| self.elasticnet_beta_0 = proximity.ElasticNet(linear.Identity(), | ||
| alpha=weights, | ||
| beta=0) | ||
| self.one_support = proximity.KSupportNorm(beta=0.2, k_value=1) |
There was a problem hiding this comment.
I might need to change the beta parameter so it basically rescales wit the ridge in case of k_value = dimension
There was a problem hiding this comment.
Are you planning to make this change in the current PR?
There was a problem hiding this comment.
Do you think it should be a great idea to have the beta parameter rescaled the same way, to have the same output as the ridge ?
There was a problem hiding this comment.
I think you are the expert in this context so I have defer to your judgement. Either way I will be happy to merge the PR.
There was a problem hiding this comment.
I'm making it compatible with the ridge, it will ease the comparison between the two methods.
|
@sfarrens I'm not sure that I can add more tests to increase the coverage what do you think? |
modopt/opt/proximity.py
Outdated
| \underset{y \in \mathbb{C}^{N}}{\text{min}} 0.5*\|x - y|\_2^2 + | ||
| \frac{\beta}{2}\text{min}\{\sum_{I \in \mathcal{G}_k}\|v_I\|_2^2: | ||
| \text{supp} (v_I) \subseteq I, \sum_{I \in \mathcal{G}_k} v_I = y\} |
There was a problem hiding this comment.
I don't think this will render without the .. math:: command.
There was a problem hiding this comment.
The expression is a bit complex, maybe we should just refer the user to the papers in references?
There was a problem hiding this comment.
That would be completely acceptable
There was a problem hiding this comment.
I think I will refer to the paper
…ordinate has not been found
LElgueddari
force-pushed
the
dev_ksup_norm
branch
from
c624358 to
c7e17a0
Compare
…was wrong on the previous version
This implements the squared k-support norm and solves #68 .
The implementation of the proximity operator is based on this paper although it has originally been proposed here