[SPARK-21340] Bring pyspark BinaryClassificationMetrics to parity with the Scala API

[mgaido91]

What changes were proposed in this pull request?

Adding all the missing methods in the pyspark API for the BinaryClassificationMetrics, ie.:

• thresholds
• roc
• pr
• fMeasureByThreshold
• precisionByThreshold
• recallByThreshold

How was this patch tested?

Tested manually.

[jakecharland]

+1 on getting this verified.

[srowen]

@jakecharland that's just an automated message asking if it's OK to test. I'll kick it off

[SparkQA]

Test build #3879 has finished for PR 18622 at commit d953bac.

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

[mgaido91]

@srowen any comment on this PR? Thanks.

[holdenk]

So while we are trying to limit the changes in mllib, we are currently exposing BinaryClassificationMetrics in the ML models so this makes sense to expose.

[holdenk]

Oh wait sorry I misread that code, it looks like we have this already wrapped behind in the Spark ML API.

Since we aren't actively working on the Spark MLlib APIs right now I probably don't see us merging this.

[mgaido91]

Thank @holdenk, should I close this PR then?

[holdenk]

That would be good yes. Let me know if you're interested take a look at the PySpark ML API since we're actively working on improving that and would be great to have more interested folks there as well :)

[mgaido91]

Thanks! Might you please resolve the JIRA as "Won't Fix" then?
Yes, of course I am interested, thanks @holdenk, if I can help and contribute I am very happy!