[SPARK-10573] IndexToString transformSchema adds output field as DoubleType

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.5.0
• Fix Version/s: 1.5.1, 1.6.0
• Component/s: ML
• Target Version/s: 1.5.1, 1.6.0
• Assignee: Nick Pritchard
• Reporter: Nick Pritchard
• Shepherd: Xiangrui Meng
• Created: 12/Sep/15 04:10
• Updated: 14/Sep/15 20:32
• Resolved: 14/Sep/15 20:28

Description

Reproducible example:

val stage = new IndexToString().setInputCol("input").setOutputCol("output")
val inSchema = StructType(Seq(StructField("input", DoubleType)))
val outSchema = stage.transformSchema(inSchema)
assert(outSchema("output").dataType == StringType)

The root cause seems to be that it uses NominalAttribute.toStructField which assumes DoubleType. It would probably be better to just use SchemaUtils.appendColumn and explicitly set the data type.

Issue Links

• links to: [Github] Pull Request #8751 (pnpritchard)