Fix #167: bump Spark version to 2.2.0

[OlivierBlanvillain]

No description provided.

[OlivierBlanvillain]

@atamborrino It looks like your recently added test breaks with Spark 2.2.0 (I see that you have a 2.1.x specific workaround, would removing that be enough?). I gave you push permissions to this branch if you want to fix it here

[OlivierBlanvillain]

Hmm it might be independent from 2.2, I got another failure due to this VectorUDT test on the joins PR: https://travis-ci.org/typelevel/frameless/builds/276202750

[atamborrino]

From https://travis-ci.org/OlivierBlanvillain/frameless/builds/276100054?utm_source=email&utm_medium=notification it looks like Scalacheck generator fails to generate sufficient valid cases for the conditional property (trainingData.numRows >= 2 && trainingData.numCols >= 1) ==>.
I will create a specific generator for non-empty matrix to avoid this case.

[codecov-io]

Codecov Report

Merging #176 into master will not change coverage.
The diff coverage is n/a.

@@          Coverage Diff          @@
##           master   #176   +/-   ##
=====================================
  Coverage      94%    94%           
=====================================
  Files          31     31           
  Lines         667    667           
  Branches        9     13    +4     
=====================================
  Hits          627    627           
  Misses         40     40

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4c04b3f...e47ae89. Read the comment docs.

[atamborrino]

@OlivierBlanvillain fixed. My matrix generator was incorrect. Should be ok now. I also adapted the test for Spark ML 2.2.0.

[OlivierBlanvillain]

Thanks for the fix, LGTM!