Github user feynmanliang commented on the pull request:
https://github.com/apache/spark/pull/7538#issuecomment-126989567
Is it still called logistic regression when there's multiple classes (I've
always called the multi class version softmax)?
In any case, I think this pr is fine since the logistic regression pipeline
stage being modified is a binary classifier. In the future, we should
definitely generalize to something like classifierSummary and
regressionSummary
On Sat, Aug 1, 2015, 5:59 PM jkbradley <[email protected]> wrote:
> @MechCoder <https://github.com/MechCoder> Just getting to this now,
> sorry! It looks quite clean. My main question is how we should handle
> multiclass LogisticRegression. Any ideas? I could imagine 2 main options:
>
> - (A) Include the union of MulticlassMetrics and
> BinaryClassificationMetrics in the summary. For multiclass data, have
> empty/bad values (or exceptions) for binary classification metrics
which do
> not make sense for multiclass.
> - (B) Create a LogisticRegressionSummary which has multiclass metrics.
> Also create a BinaryLogisticRegressionSummary which extends
> LogisticRegressionSummary and provides binary metrics.
>
> I'd prefer (B). That could be done all in this PR, or it could be split
> into 2 (with the first only adding LogisticRegressionSummary, and the
> second adding BinaryLogisticRegressionSummary).
>
> CC: @feynmanliang <https://github.com/feynmanliang>
>
> â
> Reply to this email directly or view it on GitHub
> <https://github.com/apache/spark/pull/7538#issuecomment-126970026>.
>
---
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]
