Integration Test returns some differences to expected results

[andrewrollason]

(Follow up to discussion at https://groups.google.com/forum/#!topic/actionml-user/W7Dcj1opAQo)

Consistently getting some differences to the expected engine output from ./examples/integration-test. This is on a Ubuntu 14.04 install of pio v0.9.7-aml per http://actionml.com/docs/single_machine, ElasticSearch is 1.7.5, and UR from the master branch of https://github.com/actionml/template-scala-parallel-universal-recommendation.git

The annotated diff results are at https://gist.github.com/andrewrollason/bd0d5f9a141cd881ec3829991a9de8de and appear at least similar to the expected result in most cases and none are obviously wrong (e.g. no iPhone5/6 or items being recommended as similar to self). Instances of same-valued result ordering differences have been removed.

A fresh clone from tag v0.3.0 returned the same results. Going back to v0.2.3, integration-test ran successfully with no diffs at all.

[pferrel]

I think I also answered in email. It sounds like we have not updated the "expected-results" file in the repo. UR v0.2.3 had a different query set and had a bug that would have produced different results. If you are getting consistent results ignore for now and we'll see if it's our error.