[SPARK-21388][ML][PySpark] GBTs inherit from HasStepSize & LInearSVC from HasThreshold

[zhengruifeng]

What changes were proposed in this pull request?

GBTs inherit from HasStepSize & LInearSVC/Binarizer from HasThreshold

How was this patch tested?

existing tests

[SparkQA]

Test build #79557 has finished for PR 18612 at commit 40f52cc.

• This patch fails MiMa tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
• class Binarizer(JavaTransformer, HasInputCol, HasOutputCol, HasThreshold,

[SparkQA]

Test build #79560 has finished for PR 18612 at commit 3e92a7e.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[yanboliang]

Maybe the error can be eliminated after removing this line, and actually we should set default value in the concrete class if it's inherited.

[SparkQA]

Test build #79652 has finished for PR 18612 at commit 94f6ca5.

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

[holdenk]

So improving the doc certainly makes sense on the Python side (thank you for catching that). But as @MLnick asked on the JIRA - I'm not super sure I see the point of the refactor on the Scala side. If we have a parameter with the same name but a different meaning why mix in a base trait only to override its only meaningful provided value?

What do you think @jkbradley?

[zhengruifeng]

@holdenk I think the meaning of StepSize in GBT and Threshould in LInearSVC/Binarizer is almost same as that in other algs, so it maybe better to make them inherit from same trait?

[MLnick]

It makes sense for LinearSVC since the threshold has the same meaning for other binary classifiers (e.g. LogisticRegression). The meaning of threshold for Binarizer is very different as it operates on feature values in a feature vector, not raw predicted values from a model. I would say back out the Binarizer changes here.

As for GBT, well I am sort of indifferent as the step size is roughly analogous between the two definitions (in both cases it is a learning rate).

[SparkQA]

Test build #79897 has finished for PR 18612 at commit aaa9187.

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

[MLnick]

This also needs to be reverted.

[MLnick]

and similarly here, added back

[SparkQA]

Test build #80001 has finished for PR 18612 at commit fb2ce45.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
• class Binarizer(JavaTransformer, HasInputCol, HasOutputCol, JavaMLReadable, JavaMLWritable):

[yanboliang]

LGTM

[yanboliang]

Merged into master, thanks for all.