There is some additional configuration which influences how SerializableMLibClassifier gets written/read with ObjectOutputStream/ObjectInputStream. That causes inconsitent behaviour and raises errors.
For example, name of the package "au.com.csiro.data61" / "au.com.nicta" are part of it.
I've added custom deserialization to ModelPredictor and removed fork := true in Build.scala.
This fixes the issue when running the server.
The tests for ModelRestAPI have been fixed as well.
There is some additional configuration which influences how SerializableMLibClassifier gets written/read with ObjectOutputStream/ObjectInputStream. That causes inconsitent behaviour and raises errors. For example, name of the package "au.com.csiro.data61" / "au.com.nicta" are part of it.
Schema-matcher-api does not have this issue.
Adding fork := true to the general settings fixes this problem, but raises other issues https://gist.github.com/ramn/5566596#file-serialization-scala