samoa-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From nicolas-kourtellis <>
Subject [GitHub] incubator-samoa issue #55: SAMOA-49: Added adapter for Apache Apex
Date Sat, 18 Jun 2016 17:13:31 GMT
Github user nicolas-kourtellis commented on the issue:
    Hi bhupeshchawda,
    I was checking the adapter on apex local mode and I had some initial comments:
    1) The execution of the classification on both a fixed arff file (covTypeNorm) and the
random tree generator are much slower (even 5-10 times slower) in comparison to the samoa-local
(the basic local mode of Samoa for testing purposes). Can you check if this is a matter of
the necessary overhead for the distributed execution or something else is causing the delay?
I tested with various numbers for parallelism hint, but the observation holds.
    2) The results on covtypenorm should have been deterministic. That is, given that the
file is the same used in repetitive runs, the results should be identical across runs (for
sure given that parallelism is 1). However, when I run the VHT repetitively on the same arff
file, the results are not the same across runs. Is there any fundamental reason why this would
be happening? (Again, you can check for this using the local mode for testing).
    3) The results with the random tree generator show a good accuracy and similar to the
one received with basic local mode of Samoa for testing. But the results on the arff file
degrade in performance, lower than the basic local mode of Samoa. Any reason why this is happening?
Maybe it has to do with the problem in comment 2?
    4) I noticed there are some changes in your PR regarding the samoa-api in about 10 files.
These changes seem to be mostly related to Kryo serialization. Can you explain why you needed
these done now? Also, do they affect the execution with the other adaptors and DSPEs? Also,
maybe it would have made sense to push them in a different PR? Or do they need to be there
for Apex to play along in the first place?

If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at or file a JIRA ticket
with INFRA.

View raw message