[SPARK-8291] [MLlib] [PySpark] Add parse functionality to LabeledPoint in PySpark

[MechCoder]

It is useful to have functionality that can parse a string into a LabeledPoint while loading files, etc

[MechCoder]

cc: @srowen @brkyvz Can you please have a look at this?

It will be useful in this example. https://github.com/apache/spark/pull/6744/files#diff-c433cb6c11f49d555c3741b41ffb7eecR790

[SparkQA]

Test build #34600 has finished for PR 6746 at commit 6fd40a8.

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

[MechCoder]

@davies Could you have a look at this also? :P

[davies]

There could be many different formats in a string, so I'm not sure which format we should support. It's better to be done by user.

cc @mengxr

[MechCoder]

Hmm. The supported format in this PR is coherent with that done in Scala.

[davies]

I see, in order to have similar functionality as Scala, we should do similar things as in #685, having MLUtils.loadLabelPoints() ...

[davies]

Also the format we used here, should match with LabelPoint.str(), then it can load the RDD after saveAsTextFile.

[MechCoder]

I have added tests to verify it.

from pyspark.mllib.regression import LabeledPoint
lb = LabeledPoint(2, [0.1, 1.2, 3.4])
rdd = sc.parallelize([lb, lb, lb])
rdd.saveAsTextFile("tmp")
sc.textFile("tmp").map(LabeledPoint.parse)

[MechCoder]

MLUtils.LabeledPoints is just a wrapper around the scala code right? Were you trying to infer that this also should be a wrapper around the parse method in Scala? Is that required for something as simple as this?

[MechCoder]

@mengxr Should I close this then?