[
https://issues.apache.org/jira/browse/MAHOUT-1660?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14581154#comment-14581154
]
ASF GitHub Bot commented on MAHOUT-1660:
----------------------------------------
Github user dlyubimov commented on the pull request:
https://github.com/apache/mahout/pull/135#issuecomment-110933266
yeah.. although this might be tough, it doesn't look terribly likely that
we want to introduce a lot. We had a conversation about it with @pferrel who
added a bunch of ??? stuff to H20Engine which do not seem terribly relevant to
algebraic stuff (to me). We agreed that if indexed format is needed, perhaps
this is something to solve ouside of individual backend scope. Either way,
something pending as ??? forever is not the effect we want -- and i suggest to
apply a higher level of scrutiny on what is going on there.
in my case, motivation of adding (replacing) dedicated physical operator
computing elementwise scalar operation with elementwise function operator is
pretty clear. it allows to detect and optimize and fuse more general
elementwise cases on upper level of optimizations, such as for example 1 + drmP
* dlog(drmP), into one physical map. Also it detects some very important and
frequent enough cases like X * X (self-hadamard) and sends it into this unary
elementwise function instead of AewB.
But in case of indexed dictionary structure i am not sure i have heard
convincing arguments this needs to be there... at least in the algebraic engine
itself... i am perfectly fine with quasi-algebraic support with what @pferrel
is doing. I can quite imagine this will not be an outlier case...
> Hadoop1HDFSUtil.readDRMHEader should be taking Hadoop conf
> ----------------------------------------------------------
>
> Key: MAHOUT-1660
> URL: https://issues.apache.org/jira/browse/MAHOUT-1660
> Project: Mahout
> Issue Type: Bug
> Components: spark
> Affects Versions: 0.10.0
> Reporter: Suneel Marthi
> Assignee: Dmitriy Lyubimov
> Priority: Minor
> Fix For: 0.10.2
>
>
> Hadoop1HDFSUtil.readDRMHEader should be taking Hadoop configuration from
> Context and not ignore it
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
