[SPARK-8774] [ML] Add R model formula with basic support as a transformer

[ericl]

This implements minimal R formula support as a feature transformer. Both numeric and string labels are supported, but features must be numeric for now.

cc @mengxr

[SparkQA]

Test build #37166 has finished for PR 7381 at commit fb0826b.

• This patch fails Scala style tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #37170 has finished for PR 7381 at commit 5765ec6.

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

[SparkQA]

Test build #37167 has finished for PR 7381 at commit 1f361b0.

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

[mengxr]

Also mention the operators we support and a link to the official R formula document.

[ericl]

Done

[SparkQA]

Test build #37282 has finished for PR 7381 at commit dc3c943.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • class RModelFormula(override val uid: String)

[mengxr]

Use http://stat.ethz.ch/R-manual/R-patched/library/stats/html/formula.html instead, which is in the raw R manual format.

[mengxr]

remove unused imports

[mengxr]

@ericl I make another pass. The major issue is actually that RModelFormula should be an Estimator instead of a Transformer in order to handle String columns. It requires some changes to the current implementation. So I would suggest removing the support for string labels in this PR and address it in a follow-up PR, since we already reviewed most of the code. It is okay to just comment out the test. Does it sound good to you?

[ericl]

@mengxr That makes sense, I'll do that in a followup PR. I also addressed the comments.

[SparkQA]

Test build #37425 has finished for PR 7381 at commit 2db68aa.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • class RFormula(override val uid: String)

[SparkQA]

Test build #37426 has finished for PR 7381 at commit d1959d2.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • class RFormula(override val uid: String)

[mengxr]

LGTM except some minor comments, which we can fix in the next PR. Merged into master. Thanks! As the next step, we can create a wrapper for RFormula + LinearRegression on the Scala side and then call it in R. Independently, we can add features to RModelParser. I'd recommend the former first in order to have some working MLlib + SparkR features in 1.5.

[ericl]

Sounds good, I'll look at the R integration next.