Re: a couple of swarm questions

[Rick Riolo]

re Step, I was just thinking of "run one Time step."
I agree with ginger, run N steps is very common, too.
(I can see your point about other "step" sizes, but I'll
bet one T step will have a lot of use.)

Re: the forthcoming better starting of swarm, including batch mode:
that's good news.  (I've been doing it via tcl scripts
outside of swarm, to do multiple runs with different seeds,
to sweep parameters, etc.)

In that regard, (perhaps I mentioned this before...) one feature 
that I think is important is to have each run produce a 
file of parameter settings, including the RNG seed, 
that can be easily fed back into a subsequent run
(eg thru a command line parameter specifying a file of
parameter settings) to reproduce that run exactly.  
So perhaps with the Experiment object you mentioned
below, an easy way to save it and then restart a run using
those same values.
It will also be nice to have an easy way to re-produce
a set of runs, all with the same parameter settings,
but each with different seeds (but the same seeds as
used the first time the runs were done).
And finally, to be able reproduce a parameter sweep
of runs, each with a bunch of runs with unique seeds,
but again using the same set of seeds as when the first
runs were done.

thanks.
 - r

Rick Riolo                       address@hidden
Program for Study of Complex Systems (PSCS)
1061 Randall Lab     University of Michigan
Ann Arbor MI 48109-1120
http://pscs.physics.lsa.umich.edu/rlr-home.html

On Fri, 26 Apr 1996, Nelson Minar wrote:

> Date: Fri, 26 Apr 96 10:04:47 MDT
> From: Nelson Minar <address@hidden>
> To: address@hidden
> Cc: address@hidden
> Subject: Re: a couple of swarm questions
> 
> >0. Just between you and me (just whisper back)...is a new
> >   version of swarm about to be released?
> >   Will it have some batch controls mechanisms in it?
> 
> yes, and yes. I'm hoping to have one more incremental release before
> we all go to Alife V May 13, and then another release for Alife V
> itself. My big contribution is a new way of starting up Swarm,
> including batch mode and better control of how the thing runs. It's
> shaping up pretty well, still more work to do.
> 
> >1. re the tk sim control widget the demos use:
> >   it would be nice to have a Step button to go along with run.
> 
> What do you expect step to do exactly? My best guess is "run the next
> pending action", whether it be an atomic action or a foreach. Another
> reasonable option is "run to the next time", at least if you're
> running a model that has time. I could think of other things step
> could mean, too :-) Clearly some better control is needed.
> 
> >2. re the tk parameter setting widget the demos use:
> >   it would be nice if it didn't get linked directly
> >   to addresses of parameters, but instead got linked
> >   to methods.
> 
> As part of the batch overhaul, the "form" thing has been replaced with
> a probe object on a special Experiment object which encapsulates all
> your variables. Much less hackish, and I believe it should be possible
> to have methods as well as raw variables.
> 
> >3. re probes:  maybe they can do this already.
> >   It would be nice if, when a probe is made into
> >   an agent, the probe would send a special
> >   "initializeForProbe" message to the agent
> 
> I think Manor will have to answer this - I believe something like this
> is doable, if nothing else than by overriding giveCompleteProbeDisplay.
>