twitter_classifier/Collect.scala: Pending TODO can be completed: SPARK-3390 was fixed in Spark 1.2.0.

[MiguelPeralvo]

Line 42 of reference-apps/twitter_classifier/scala/src/main/scala/com/databricks/apps/twitter_classifier/Collect.scala can now be safely removed, as SPARK-3390 was fixed in pull request #2364 for Apache 1.2.0.

If you use Spark 1.2.0, this is the code that can be removed:

.filter(!_.contains("boundingBoxCoordinates")) // TODO(vida): Remove this workaround when SPARK-3390 is fixed.

If you remove it for Spark 1.1.0, Collect.java won't break when run, but ExamineAndTrain.scala will do, with a "scala.MatchError: StructType(List())" exception. It will be caused by the "boundingBoxCoordinates" json entries, as Spark 1.1.0 doesn't handle them properly.