Description
Under spark.mllib, if number blocks is set to -1, we set the block size automatically based on the input partition size. However, this behavior is not preserved in the spark.ml API. If user sets -1 in Spark 1.3, it will not work, but no error messages will show.