[ 
https://issues.apache.org/jira/browse/FLINK-3771?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15258288#comment-15258288
 ] 

ASF GitHub Bot commented on FLINK-3771:
---------------------------------------

Github user greghogan commented on the pull request:

    https://github.com/apache/flink/pull/1900#issuecomment-214782976
  
    Hi @vasia,
    Comparing the APIs, `Translate` processes at the level of label or value 
whereas `mapVertices/Edges` process vertices and edges. What are use cases for 
the extra context (namely, the label) provided by `mapVertices/Edges`? 
Implementations of `Translator` can be reused across functions whereas the 
`MapFunction` provided to `mapVertices/Edges` are specific to that method. 
Would `mapLabels` require two new `MapFunction` or use an interface like 
`Translator`?
    
    `Translator` is wrapped by a specific `MapFunction` for each `Translate` 
function.
    
    Only one of the five current `Translate` functions operates on a `Graph`. 
`tranlateGraphLabels` could be moved to `Graph`, and there could be a 
`translateVertexValues` and `translateEdgeValues` also added to `Graph`. We 
don't always have a `Graph`, sometimes algorithms return a `DataSet` of 
vertices or edges, so having the static methods is helpful for these 
specialized `Vertex` and `Edge` operations.


> Methods for translating Graphs
> ------------------------------
>
>                 Key: FLINK-3771
>                 URL: https://issues.apache.org/jira/browse/FLINK-3771
>             Project: Flink
>          Issue Type: New Feature
>          Components: Gelly
>    Affects Versions: 1.1.0
>            Reporter: Greg Hogan
>            Assignee: Greg Hogan
>             Fix For: 1.1.0
>
>
> Provide methods for translation of the type or value of graph labels, vertex 
> values, and edge values.
> Sample use cases:
> * shifting graph labels in order to union generated graphs or graphs read 
> from multiple sources
> * downsizing labels or values since algorithms prefer to generate wide types 
> which may be expensive for further computation
> * changing label type for testing or benchmarking alternative code paths



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to