shogun lars,lasso fixes #15
Conversation
| ['datasets/madelon_X.csv', 'datasets/madelon_y.csv']] | ||
| options: '-l 0.01' | ||
|
|
||
| LARS: |
There was a problem hiding this comment.
Is there reason this was left out? methods/shogun/lars.py was already present , so curious.
|
@karlnapf some fixes for LARS benchmarks input, which lead to wrong results. |
| model.train(RealFeatures(X.T)) | ||
| model.set_labels(RegressionLabels(responsesData)) | ||
| model.train(RealFeatures(inputData.T)) | ||
|
|
There was a problem hiding this comment.
Comparing with scikit lasso doesnt make sense actually since its uses coordinate descent to solve as opposed to lars.
There was a problem hiding this comment.
Not sure I get this.
Definitely, it only makes sense to compare against this one
There was a problem hiding this comment.
Ok I just checked. This should not be merged.
@Saurabh7 you are right that comparing sklearn's LASSO with Shogun's LARS doesnt make any sense.
Instead, can you do the right comparison of the various LARS implementations?
|
Can one of the admins verify this patch? |
|
This is old stuff |
3 participants
No description provided.