On Thu, May 2, 2013 at 5:07 PM, Andreas Mueller <[email protected]>wrote:
> Cool.
> How does the performance differ from the sklearn implementation?
>
I would be interested to know too. Basically lightning contains alternative
implementations for:
- Lasso / MultitaskLasso [in primal_cd.CDRegressor]
- SGDClassifier / SGDRegressor [in sgd.SGDClassifier / sgd.SGDRegressor]
- LinearSVC(dual=True, penalty="l2"), LinearSVC(dual=False,
penalty="l2|l1") [in dual_cd.LinearSVC and primal_cd.CDClassifier]
- LogisticRegression(penalty="l1|l2") [in primal_cd.CDClassifier]
+ other stuff not in scikit-learn.
Compared to liblinear, it would be interesting to know what's the penalty
of the dataset abstraction (due to virtual method calls). I would guess
that lightning could be faster for dense data or large-scale sparse
datasets, since it works directly with the native numpy / scipy data
structures.
> And when will we see the PR ;)
>
No immediate plan as I'm busy with starting a new job and my new research
:) But lightning follows scikit-learn's API conventions so it should be
easy to use for anyone knowing scikit-learn.
Mathieu
------------------------------------------------------------------------------
Introducing AppDynamics Lite, a free troubleshooting tool for Java/.NET
Get 100% visibility into your production application - at no cost.
Code-level diagnostics for performance bottlenecks with <2% overhead
Download for free and get started troubleshooting in minutes.
http://p.sf.net/sfu/appdyn_d2d_ap1
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
