[jboss-jira] [JBoss JIRA] Commented: (JBRULES-2900) Consider concurrency tasks optimization (both Jacoby and Gauss-Seidel iterations)

[Geoffrey De Smet (JIRA)]

    [ https://issues.jboss.org/browse/JBRULES-2900?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12587702#comment-12587702 ] 

Geoffrey De Smet commented on JBRULES-2900:
-------------------------------------------

This should be an interesting experiment.

Personally I have my doubts:
Both the Jacoby or Gauss-Seidel method are based on writing the constraints as mathematical equitations.
In Drools Planner, constraints are written as the far more flexible rule engine rules (DRL),
which supports Object Oriented Programming, re-use of java code (for example HolidayHelper.isHoliday(date, country)), complex patterns (correlation, logical insertions, ...), custom accumulate functions, ...
However, I doubt this allows for Jacoby/Gauss-Seidel.
Furthermore, in my experience with the planning competitions, I haven't seen exact methods such as simplex, the Hungarian method, Jacoy, Gauss-Seidel been able to scale to the size of data and constraints in the real world.


That being said, I can be proven wrong.
Pull requests (patches) are very welcome on the source code:
  https://github.com/droolsjbpm/drools-planner
If implemented, it should be interesting to compare results on the already implement examples, such as TTP and examination.

> Consider concurrency tasks optimization (both Jacoby and Gauss-Seidel iterations)
> ---------------------------------------------------------------------------------
>
>                 Key: JBRULES-2900
>                 URL: https://issues.jboss.org/browse/JBRULES-2900
>             Project: Drools
>          Issue Type: Feature Request
>      Security Level: Public(Everyone can see) 
>          Components: drools-planner
>    Affects Versions: FUTURE
>            Reporter: Michał Warecki
>            Assignee: Geoffrey De Smet
>            Priority: Minor
>             Fix For: FUTURE
>
>
> Will be great to support concurrency tasks optimization. 
> In example we have 3 tasks A, B, C and objective function 2A^2 + A*B + B^2 + AC. With Gauss-Saidel iteration (where task B in time+2 depends on A in time+1 and C in time+2 depends on A in time+1) we can execute tasks A,B,C in order: A and BC cocurrently, so in 2 iterations. Without optimization tasks will be executed in 3 iterations (A , B, C).
> We should support Jocoby iteration as well where tasks B and C does not depend on A in previous iteration.
> This can be done with Gradient Method.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira