systemml-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alok Singh (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SYSTEMML-1542) SystemML shouldn't enforce write statement need to be defined in all the if else branch
Date Tue, 25 Apr 2017 22:21:04 GMT

    [ https://issues.apache.org/jira/browse/SYSTEMML-1542?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15983734#comment-15983734
] 

Alok Singh commented on SYSTEMML-1542:
--------------------------------------

Here are the couple of solutions from the brainstorming session. 

1) Have the systemML returns NA_matrix . the concept of NA matrix would be similar to SQL
like we have sparse matrix, dense matrix, empty matrix or NA_matrix . empty matrix return
is also fine

2) user rewrite the DML so that there is no branch of if else and write statement. But I feel
this is not the right solution since whatif there is the nested functions and write is in
that nested function. it would be too much work for the end user to modify

3) parse the error message from dml and make appropriate decision in the downstream apps.
However, sometimes systemML err msg can be very convulated and also I think having end user
to change it is not good

4) we give stern warning saying that all the write are not there in all branches and we can
also have a flag which when set will allow systemML to return empty matrix in case of no matrix
in write in the if else clause.



> SystemML shouldn't enforce write statement need to be defined in all the if else branch
> ---------------------------------------------------------------------------------------
>
>                 Key: SYSTEMML-1542
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-1542
>             Project: SystemML
>          Issue Type: Story
>            Reporter: Alok Singh
>
> It seems like we have to define all the write cmd in all the possible systemML if else
branch. 
> It could become very complicated for the nested function call to take this into account.
> I think systemML should automatically insert write(file, empty_matrix) if the all the
branches of the DAG doesn't contains the write statement
> Note that in reference to the mlCtx, this pb manifest itself into big stack trace
> Note that this bug is also related to https://issues.apache.org/jira/browse/SYSTEMML-1422



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message