Github user shivaram commented on the pull request:
https://github.com/apache/spark/pull/3637#issuecomment-66403895
@jkbradley Apologies for the delay - I just read your design doc and am
catching up on this discussion.
Sorry if I missed something, but could you clarify the use case here ? I
can see two kinds of scenarios
1. Cases where we just want to use existing classifier like
LogisticRegression in a pipeline. I guess the train() interface shouldn't
really matter here as they we will be passing around SchemaRDDs in a pipeline
and calling fit (thus going through the untyped API ?).
2. Cases where developers want to implement a new Classification or
Regression method. For these cases I think the strongly typed API would help in
reducing the amount of cruft code and possible bugs in extracting features,
labels etc.
FWIW I agree with the conclusion of keeping LabeledPoint simple as (Double,
Array[Double]). And I think `predictRaw` is also probably fine as meaning of
the values returned may vary (as noted in your comment).
---
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]
