RE: Is this the defects of Javaswarm? the swarm kernel neglect new added agents!

[Hu, Jianjun]

Thanks Paul.

I think one solution maybe swarm kernel exposes its iterator to the user,
then we can use this iterator to update the agent list.
and then the swarm kernel can track the changes of agents.

Jianjun Hu




-----Original Message-----
From: address@hidden
[mailto:address@hidden Behalf Of Paul Johnson
Sent: Wednesday, June 26, 2002 10:32 AM
To: swarm-support
Cc: Hu, Jianjun
Subject: Re: Is this the defects of Javaswarm? the swarm kernel neglect new
added agents!


I just noticed this letter from March in my Inbox, and I don't know that
anybody has answered it.

To tell the truth, lately I have been circumventing this whole issue of
createFActionForEach and such by just scheduing a method in ModelSwarm
that simply iterates over the list. In my JelFarol (java El Farol)
example (its in the ftp apps/java/contrib) I do it like this.

Here's a method that steps through the list

     public void stepAgents (){
         int x;
         Iterator iterator = agents.listIterator(0);
         Agent anAgent = null ;

         while (iterator.hasNext()){
             anAgent = (Agent) iterator.next();
             x = anAgent.makeChoice();
         }
      }



In buildActions, I have this:

    ActionGroup modelActions = new ActionGroupImpl(getZone());

   try {
             modelActions.createActionTo$message
                 (this, new Selector (this.getClass (), "stepAgents",
false));
         } catch (Exception e) {
             e.printStackTrace (System.err);
             System.exit (1);
         }


         modelSchedule = new ScheduleImpl (getZone(), 1);
         modelSchedule.at$createAction (0, modelActions);


}

This approach works in a limited situation where you want all agents
processed as a group in a certain section of time, and you don't want
other actions to be thrown into the schedule in between your agents.
This approach certainly does not work if you have 10 lists of agents and
you want actions for them to be thrown onto the schedule at time X and
then you expect Swarm to randomize them.  But for one list, it makes
sure they are handled.

This way is simpler and generally faster than the createFActionForEach
regime.  It is faster because it does not require the swarm scheduling
system to do any work for each individual agent. All one must do is
schedule a trip through the list.

I've been thinking a while, and I know a few people agree with me, that
somebody ought to invest a lot of personal effort in writing down the
convenient ways of scheduling in Swarm-2.1+.  Marcus has provided useful
notes here and there, but there still remains a pretty big translation
job to synthesize it into a FAQ or User Guide.  That is something that
will probably get done as soon as I finish some other Swarm related jobs
and member recruitment.


> Hu, Jianjun wrote:
>
>>Help!
>>
>>I am running difficulty into the javaswarm when I want to remove the agent
>>from the agent  list.  I just can't remove the dead agent from the list.
>>In addition, when I add some new agent to the list,  these agents just be
>>neglected by the swarm kernel and won't get executed.
>>I am using swarm2.1.1 java version.
>>
>>I guess the problem is that the javaswarm use the iterator to trace the
>>agent list when I register this list in buildaction. but later when I
>
> update
>
>>the agent list by removing or adding agents, the iterator that the kernel
>>keeps just fails and become invalid.
>>
>>What's the solution or other possible reason?
>>
>>
>>
>>In my  doveModelSwarm class's buildobject function, I add the intial list
>
> of
>
>>agent just like heatbugs
>>              tadpoleLayer.agentList =  new ArrayList();
>>
>>
>>              //create the tadpoles
>>              for( i=0;i<nTadples;i++){
>>              QTadpole newTadpole = new QTadpole(tadpoleLayer,this);
>>              tadpoleLayer.agentList.add(newTadpole );
>>              }
>>In buildaction function:  I register the agentlist for  simulation
>>
>>    try {
>>      QTadpole proto = (QTadpole) tadpoleLayer.getAgentList().get (0);
>>      Selector sel =
>>        new Selector (proto.getClass (), "update", false);
>>      actionForEach =
>>        modelActions.createFActionForEachHomogeneous$call
>>        (tadpoleLayer.getAgentList(),
>>         new FCallImpl (this, proto, sel,
>>                        new FArgumentsImpl (this, sel, true)));
>>    } catch (Exception e) {
>>      e.printStackTrace (System.err);
>>    }
>>
>>But after some steps, when I add new agent by
>>agentList.add(newAgent);
>>these agents are  just neglected by the kernel.
>>
>>when I use agentList.remove(oldAgent).
>>This old agent can't be removed.
>>
>>
>>hmm????
>>
>>
>>Thank you very much
>>
>>Jianjun Hu
>>

> --
> Paul E. Johnson                       email: address@hidden
> Dept. of Political Science            http://lark.cc.ku.edu/~pauljohn
> University of Kansas                  Office: (785) 864-9086
> Lawrence, Kansas 66045                FAX: (785) 864-5700
>



--
Paul E. Johnson                       email: address@hidden
Dept. of Political Science            http://lark.cc.ku.edu/~pauljohn
University of Kansas                  Office: (785) 864-9086
Lawrence, Kansas 66045                FAX: (785) 864-5700


                  ==================================
   Swarm-Support is for discussion of the technical details of the day
   to day usage of Swarm.  For list administration needs (esp.
   [un]subscribing), please send a message to <address@hidden>
   with "help" in the body of the message.



                  ==================================
   Swarm-Support is for discussion of the technical details of the day
   to day usage of Swarm.  For list administration needs (esp.
   [un]subscribing), please send a message to <address@hidden>
   with "help" in the body of the message.