Github user yanboliang commented on the pull request:
https://github.com/apache/spark/pull/11247#issuecomment-186040385
@dbtsai Thanks for clarification, I understand the consideration of the
current implementation.
I know that if features are not standardized at all, it may result
convergency issue when the features have very different scales. This is aligned
with setting ```standardization = true``` as default value.
But to some special case(such as the test case which I mentioned above),
disable standardization will accelerate training. I think the users should have
the ability to control it.
Further more, I think enable or disable ```standardization``` will run the
same route may make users confused. Users may look forward to different result
or convergency rate(even if it's worse in some cases) for enable or disable
```standardization```, but they got the same one.
The only issue we should concern about is that if this change will break
someone's working training. I think most of the working trainings may trained
with regularization, it will has little effect. Even if users train without
regularization, enable or disable ```standardization``` will produce the same
result currently, users have high possibility to use the default setting.
I'm still open to hear your thoughts.
BTW, I send #11258 to fix the bug for ```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]
