Hi all, I have posted my issue on the AEMR forum and there is some input from there: http://developer.amazonwebservices.com/connect/thread.jspa?threadID=32028&tstart=0 I can't work on this right now but just wanted to keep you updated. Specifically it looks like on the Mahout side, we don't need to unpack all jars into the top-level .job jar since Hadoop is quite happy with jars inside jars (which puzzles me, since the Java classloader can't load these -- if someone understands what is going on I'm keen to know).
Sebastien
