subclass of TypedCompositeActor which, when fired,
executes the contained model much the way the top-level
Manager currently executes a model.
This sounds reasonably straightforward.
I was just talking to Steve Neuendorffer about this,
and one idea is for inputs and outputs to be provided
by port parameters. I'm not quite as clear on how to
transport the results to an output...
Since port parameters are not yet released (only
in runtime-only form), is there some reasonable
hack that you can suggest to try? If all the
parameters are transmitted via ports, this could work I
think. A single token arrives for each parameter at
each input port, the firing of the composite actor
causes the consumption of those tokens and subsequent
internal simulation with many iterations, and then a
single token is supplied to the output port(s) which
then propagates at the upper level model. It seems
that the key issue is matching the rates kind of as
described under vectorization in the design document.
Would this be practical?
I would love to know that I have some way to succeed.
If you give me a rough outline of the steps, I would be
happy to give it a try and contribute back the code.
Again thanks - Jim
Edward A Lee wrote:
I've been wanting to do something like this for some time... The way that I would do it would be to create a subclass of TypedCompositeActor which, when fired, executes the contained model much the way the top-level Manager currently executes a model.What I'm not quite so clear on is how to get initial configuration data in, and how to get execution results out. I was just talking to Steve Neuendorffer about this, and one idea is for inputs and outputs to be provided by port parameters. I'm not quite as clear on how to transport the results to an output... Edward At 11:10 AM 2/14/2003 -0800, Jim Kleckner wrote:I have dipped into the descriptions of the various simulation domains and haven't found anything quite like what I'm looking for. Again, apologies if I have overlooked an obvious source. Section 2.5 of this document: http://ptolemy.eecs.berkeley.edu/ptolemyII/hyvisual2.2/index.htm has some interesting mixed model work but seems to still constrain the contained models to be consistent with a global time line. Is that the case? The ideal description of what I would like to do is to take a simulation model and embed it into another model that lets me manipulate the parameters. To be concrete, let's say that I have a DE domain model M that takes some time sequence data and computes a scalar measure f as a result of the completed execution of the model M. At first, I would just like to take the slider class and connect it to one or more of the parameters, say x, of model M and plot the sequence f(x) in a sequence plotter. Each value in f(x) is the result of initializing and sweeping of time over the model. The second step would be to drive a set of parameters x,y,z with input distributions and to get a plot of the resulting distribution of f(x,y,z). The next step after that would be to perform some sort of optimization over f. Is this possible with Ptolemy and what would be the best way to approach this? I suppose one could build a director with explicit knowledge of a contained director. One might also envision some kind of construct analogous to the "PortParameter", say "DirectorParameter", which gave a structural way to poke the control functions of the director. Thanks - Jim
---------------------------------------------------------------------------- Posted to the ptolemy-hackers mailing list. Please send administrative mail for this list to: [EMAIL PROTECTED]