[
https://issues.apache.org/jira/browse/SPARK-13944?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15232006#comment-15232006
]
Sean Owen commented on SPARK-13944:
-----------------------------------
I can understand the idea of being able to use the simple API classes (vectors,
models, etc) as a separable module. However, the theory behind classes like
{{Vector}} was that they're just Spark-specific wrappers, and not something
you'd use outside the context of a Spark app. The model classes likewise depend
on Spark classes like {{RDD}} since they're primarily there to score
Spark-specific representations of data.
For general interchange, I'd assume one would use general representations --
PMML, JSON, etc. And then for scoring not related to Spark, use libraries that
already exist for this purpose like JPMML.
I still understand that, well, it seems funny to have this Spark
NaiveBayesModel that can score a single input vector but tell people they can't
use it unless they drag spark-core into their app. It does mean you're pushing
Spark towards also becoming a general non-distributed model representation and
scoring library. Right now, it isn't that, and taking steps to advertise its
half-baked state as such seem problematic. Same with PMML support -- seems like
it's better than nothing, but supporting a little PMML has turned out to be
almost worse than not at all.
Maybe I'm over-thinking this and this is mostly about the {{.linalg}} classes?
while it's kind of unfortunate that these internal wrapper classes have
"leaked", that cat may be out of the bag. I recognize that's already a problem
in that you have to depend on all of mllib to use VectorUDT for example.
> Separate out local linear algebra as a standalone module without Spark
> dependency
> ---------------------------------------------------------------------------------
>
> Key: SPARK-13944
> URL: https://issues.apache.org/jira/browse/SPARK-13944
> Project: Spark
> Issue Type: New Feature
> Components: Build, ML
> Affects Versions: 2.0.0
> Reporter: Xiangrui Meng
> Assignee: DB Tsai
> Priority: Blocker
>
> Separate out linear algebra as a standalone module without Spark dependency
> to simplify production deployment. We can call the new module
> spark-mllib-local, which might contain local models in the future.
> The major issue is to remove dependencies on user-defined types.
> The package name will be changed from mllib to ml. For example, Vector will
> be changed from `org.apache.spark.mllib.linalg.Vector` to
> `org.apache.spark.ml.linalg.Vector`. The return vector type in the new ML
> pipeline will be the one in ML package; however, the existing mllib code will
> not be touched. As a result, this will potentially break the API. Also, when
> the vector is loaded from mllib vector by Spark SQL, the vector will
> automatically converted into the one in ml package.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
