[SPARK-12016][MLlib][PySpark] Wrap Word2VecModel when loading it in pyspark

[viirya]

JIRA: https://issues.apache.org/jira/browse/SPARK-12016

We should not directly use Word2VecModel in pyspark. We need to wrap it in a Word2VecModelWrapper when loading it in pyspark.

[SparkQA]

Test build #47056 has finished for PR 10100 at commit 3ba2a10.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #47061 has finished for PR 10100 at commit 6ac03ac.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[holdenk]

This is only a suggestion, but looking at the other Java loadable models which are also wrapped in python it seems the standard approach is to instead add a load[X] method to the PythonMLLibAPI instead of doing the wrapper in Python. Might be worth doing that way just for consistencies sake - what are your thoughts?

[viirya]

It makes sense. Thanks. I will update this later.

[SparkQA]

Test build #47122 has finished for PR 10100 at commit 56c250e.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[viirya]

ping @davies@mengxr

[viirya]

ping @davies Can you take a look of this? It should be straightforward one.

[davies]

LGTM, merging this into master, thanks!

[jkbradley]

I just backported this to 1.6 and 1.5