Github user tanyinyan commented on the pull request:
https://github.com/apache/spark/pull/5055#issuecomment-87147135
Hi @jkbradley , @srowen , I'm considering "option 2" these days, and look
up what libsvm and liblinear does . And find that, both in libsvm and liblinear
, scaling changes the performance.
In libsvm guide:http://www.csie.ntu.edu.tw/~cjlin/papers/guide/guide.pdf ,"
We recommend linearly
scaling each attribute to the range [â1, +1] or [0, 1]", in the appendix
A,there are some examples that scaling improve the accuracy.
The same as in
liblinear:http://cran.r-project.org/web/packages/LiblineaR/LiblineaR.pdf ,
"Classification models usually perform better if each dimension of the data is
first centered and scaled."
So, I suggest "setFeatureScaling(true)" (don't expose it as an option) as
what is done in class LogisticRegressionWithLBFGS
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
