On 22 June 2012 06:31, Ted Dunning <[email protected]> wrote: > I find a better idea is to use mvn install to put a version into your local > repo, then compile the code that uses Mahout down to a self contained jar > file and carry that to your target system. Maven will suck Mahout into > your application's jar file from your local repo and you will be all set.
Sounds useful in general, although in my case I'm usually getting by with just the 'bin/mahout' commandline interface. The customisation tends more to be fixes, patch testing etc. Dan > On Thu, Jun 21, 2012 at 1:51 PM, Dan Brickley <[email protected]> wrote: > >> When I need to use Mahout on a real Hadoop cluster rather than on my >> longsuffering laptop, I use systems which are not well set up for >> compiling Mahout from source - i.e. they lack Maven, Subversion etc. >> >> On a good day, this is fine, I can just unpack >> mahout-distribution-0.7.tar.gz or whatever and use a standard release. >> But when I want to try a custom version, with my own fixes/experiments >> or with patches applied, the ideal workflow isn't obvious to me from >> https://cwiki.apache.org/MAHOUT/buildingmahout.html (and >> https://cwiki.apache.org/MAHOUT/how-to-release.html wasn't obviously >> applicable either; the mvn commands there don't like local >> modifications, amongst other things). >> >> Previously I've tried uploading jars built with 'mvn package', >> alongside bin/mahout, or zipping up the whole thing by hand. >> Presumably the answer is obvious to those who spend more time in >> Maven. I see 'mvn package' produces jars but no distribution .tar.gz. >> What is everyone else doing?
