sparse-ho
sparse-ho copied to clipboard
QB3
Fast hyperparameter settings for non-smooth estimators:
https://app.circleci.com/jobs/github/QB3/sparse-ho/1034?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link
+ harmonizing first-order methods in examples
It's the new default name from github and many repos have transitioned already. OK @QB3 @Klopfe ?
Hi @QB3 I was running Sparse-HO for rcv1_train with gradient descent and found a very strange bug. The first iteration would be OK, but then gradients would blow up and...
Currently the name of the classes encapsulating the estimators is the same as the name of the estimators: for example Lasso is the name of the estimator from sklearn/ celer...
After writing #111 this I think Lasso, ElasticNet, WeightedLasso and Logistic regression could inherit from a common class. This would avoid a bit of duplication eg the following methods: get_beta,...
An issue to summarize tests to add now that codecov works again: Criterion: - [ ] `HeldOutSmoothedHinge` is not tested (easy) - [ ] `CrossVal.get_val_grad()` is not tested - [...
There is one in test_criterion and one in common. Can we remove the one in test_criterion and define it in common ? I was trying to run only the svr...