Home » Eclipse Projects » Spatiotemporal Epidemiological Modeler (STEM) » AMP Proposal
| AMP Proposal [message #1958] |
Mon, 06 April 2009 23:15  |
 Miles Parker
Messages: 1341
Registered: July 2009 |
Senior Member |
|
|
Hi all,
Just waiting for this newsgroup to startup to introduce the STEM team
to the Agent Modeling Platform. There seems to be a lot of potential
for collaboration here -- I'm guessing that this is the only other
Eclispe project that has familiarity with complexity science and
modeling (as in modeling and simulation, not model driven development)
issues. Plus, AMP has space and time too. :)
http://eclipse.org/proposals/amp
I'm particularly interested that you seem to have done a lot of work
supporting GIS -- that is an area that AMP will eventually need direct
support for but that isn't in project plans. I'd also like to take a
look in depth at your neta-models to see where there might be
connections. Another large area directly related to STEM is exploration
of automated extraction of executable models from ontologies and biomed
data-sources. (Feel free to send me an email offline as well.)
On the AMP side, we've got a very general high-level meta-model
representation for agent behaviors that I'd love to get your thoughts
on. I've also done a lot of work on GUI issues, i.e. generation and
runtime support, UI, EMF.Edit, etc.. so perhaps there are some pieces
that you will find useful there. The AMP execution framework is planned
to provide a lot of model execution time support, including views, GEF
EditParts, etc., and significant parts of this are working now.
best regards and congratulations,
Miles
Miles Parker
Metascape, LLC
http://miles@metascapeabm.com
http://milesparker.blogspot.com
|
|
|
| Re: AMP Proposal [message #2013 is a reply to message #1958] |
Wed, 08 April 2009 13:29  |
Daniel Ford
Messages: 148
Registered: July 2009
Location: New York |
Senior Member |
|
|
I read the proposal, looks cool. We should definitely look for synergy.
The STEM geographic data sets are pretty extensive and have gone through the
all important and challenging Ecliopse IP CQ process. There's also scope
for interaction as it would be possible to create a kind of hybrid model in
which particular labels in the graph maintained by STEM were updated by
particular Agent based models.
--
Daniel Ford
IBM Almaden Research Center
San Jose, CA
"Miles Parker" <milesparker@gmail.com> wrote in message
news:gre2f2$j6c$1@build.eclipse.org...
> Hi all,
>
> Just waiting for this newsgroup to startup to introduce the STEM team to
> the Agent Modeling Platform. There seems to be a lot of potential for
> collaboration here -- I'm guessing that this is the only other Eclispe
> project that has familiarity with complexity science and modeling (as in
> modeling and simulation, not model driven development) issues. Plus, AMP
> has space and time too. :)
>
> http://eclipse.org/proposals/amp
>
> I'm particularly interested that you seem to have done a lot of work
> supporting GIS -- that is an area that AMP will eventually need direct
> support for but that isn't in project plans. I'd also like to take a look
> in depth at your neta-models to see where there might be connections.
> Another large area directly related to STEM is exploration of automated
> extraction of executable models from ontologies and biomed data-sources.
> (Feel free to send me an email offline as well.)
>
> On the AMP side, we've got a very general high-level meta-model
> representation for agent behaviors that I'd love to get your thoughts on.
> I've also done a lot of work on GUI issues, i.e. generation and runtime
> support, UI, EMF.Edit, etc.. so perhaps there are some pieces that you
> will find useful there. The AMP execution framework is planned to provide
> a lot of model execution time support, including views, GEF EditParts,
> etc., and significant parts of this are working now.
>
> best regards and congratulations,
>
> Miles
>
> Miles Parker
> Metascape, LLC
> http://miles@metascapeabm.com
> http://milesparker.blogspot.com
>
|
|
|
| Re: AMP Proposal, Equation / Agent approach [message #2027 is a reply to message #2013] |
Wed, 08 April 2009 18:51 |
Miles Parker
Messages: 1341
Registered: July 2009 |
Senior Member |
|
|
On 2009-04-08 06:29:16 -0700, "Daniel Ford" <daford@almaden.ibm.com> said:
> I read the proposal, looks cool. We should definitely look for synergy.
> The STEM geographic data sets are pretty extensive and have gone through the
> all important and challenging Ecliopse IP CQ process. There's also scope
> for interaction as it would be possible to create a kind of hybrid model in
> which particular labels in the graph maintained by STEM were updated by
> particular Agent based models.
Thanks Daniel!
Yes, the reason I brought that up is because I wanted to understand the
level of granularity as that would give a sense of where hybrid
interactions might work. Integrating and / or deriving agent and
equation model representationshas has been something I've been thinking
about for a long time. There are some really cool potentials here --
I've had a similar discuscion with Lyle Wallis ("interested party")
about integraton with dynamic systems models. Its funny, because when
you turn to abstract representation there is a strong sense in which
the distinction if more methodological than technological, esp. WRT
granularity. This is proably way too much to get into in depth in this
forum, but I do have some exposure ;) to and collegeuas in this area,
and could imagine a very nice demo model where say individual node
state comes from a contained agent-based representation. Something that
did a micro representation of contact, transmission, incubation,
etc..that fed into a macro model of population centers. Let me know if
you want to get into this further offline.
I'm anxious to get something available from the toolset soon so that
people can see UI execution aspects. Perhaps I'll try to get some
screencasts done. And the geostuff is really important WRT to
visualization and integration -- to have all the world geo sets
provided by the toolset directly would itself be extremely cool.
One other area that is ripe: I am finally in the midst of getting into
work with parameterization and model run (batch) control. I've done a
bit of support for this already; allowing text edited files to drive
model execution and validation / verification testing. But my intention
all along has been to have this be defined through an Ecore model. (See
the score paper referrnced in proposal.) The MetaABM nee AMF
representation already has a quite rich generic representation for
model attributes / parameters. There is no reason in the world that
this has to be taken as an ABM only construct; in fact generalization
to equation and agent representation was kind of a background design
goal. So what jumped out at me was your 0.3.0 description of support
for experiments over scenarios. If you've done this work already then
perhaps there in some obvious way to generalize / merge that
functionality.
Also, I wanted to point out that your support for triggers seems to
dock pretty directly to something that is an inherent part of the
MetaABM / AMF design. We have a rich construct for Actions, in fact, my
view of an ABM is as a set of Queries (in agent, space and time) and
Transformations (i.e. paramter settings, agent movements, graph
connection creation). Of course, a binary predicate is just a
disembodied query... An aspect of this that I think is really unique is
that these "Selections" (think SQL queries) can be chained, so for
example using just queries and transofmations you can do things like
have all agents withi some chacterisitcs search for a related agent
within n spaces, move toward that agent, create a network connection
with it, etc.. All models in the MetaABM system are deifned this way
and I'm pretty certain that all cannonical agent-based models can be.
All of the code gen and editors for this stuff works and is pretty
mature.
-Miles
|
|
|
| Re: AMP Proposal, Equation / Agent approach [message #2042 is a reply to message #2027] |
Fri, 10 April 2009 00:35 |
Daniel Ford
Messages: 148
Registered: July 2009
Location: New York |
Senior Member |
|
|
"Miles Parker" <milesparker@gmail.com> wrote in message
news:griro4$deb$1@build.eclipse.org...
> On 2009-04-08 06:29:16 -0700, "Daniel Ford" <daford@almaden.ibm.com> said:
>
>> I read the proposal, looks cool. We should definitely look for synergy.
>> The STEM geographic data sets are pretty extensive and have gone through
>> the
>> all important and challenging Ecliopse IP CQ process. There's also scope
>> for interaction as it would be possible to create a kind of hybrid model
>> in
>> which particular labels in the graph maintained by STEM were updated by
>> particular Agent based models.
>
> Thanks Daniel!
>
> Yes, the reason I brought that up is because I wanted to understand the
> level of granularity as that would give a sense of where hybrid
> interactions might work. Integrating and / or deriving agent and equation
> model representationshas has been something I've been thinking about for a
> long time. There are some really cool potentials here --
> I've had a similar discuscion with Lyle Wallis ("interested party") about
> integraton with dynamic systems models. Its funny, because when you turn
> to abstract representation there is a strong sense in which the
> distinction if more methodological than technological, esp. WRT
> granularity. This is proably way too much to get into in depth in this
> forum, but I do have some exposure ;) to and collegeuas in this area, and
> could imagine a very nice demo model where say individual node state comes
> from a contained agent-based representation. Something that did a micro
> representation of contact, transmission, incubation, etc..that fed into a
> macro model of population centers. Let me know if you want to get into
> this further offline.
Conceptually, this isn't hard to imagine, but in practice there would be
some challenges. The main one would be the synchronization. If we had a
number of agent models feeding into seperate nodes in a STEM simulation, the
simulation cycle time would be that of the slowest agent simulation. Right
now, for the entire United States at the county level (3000+ graph nodes),
the best cycle time is just a bit less than 500 milliseconds. I suspect
that the Agent based simulations will be a bit longer and if there are
multiple instances running in parallel they all need to be managed; that
would be a challenge.
>
> I'm anxious to get something available from the toolset soon so that
> people can see UI execution aspects. Perhaps I'll try to get some
> screencasts done. And the geostuff is really important WRT to
> visualization and integration -- to have all the world geo sets provided
> by the toolset directly would itself be extremely cool.
STEM can currently support three kinds of geographic visulatizatoin, it's on
internal view, Google Earth and, more a proof of implemenation, NASA
Whirlwinds.
>
> One other area that is ripe: I am finally in the midst of getting into
> work with parameterization and model run (batch) control. I've done a bit
> of support for this already; allowing text edited files to drive model
> execution and validation / verification testing. But my intention all
> along has been to have this be defined through an Ecore model. (See the
> score paper referrnced in proposal.) The MetaABM nee AMF representation
> already has a quite rich generic representation for model attributes /
> parameters. There is no reason in the world that this has to be taken as
> an ABM only construct; in fact generalization to equation and agent
> representation was kind of a background design goal. So what jumped out at
> me was your 0.3.0 description of support for experiments over scenarios.
> If you've done this work already then perhaps there in some obvious way to
> generalize / merge that functionality.
Yes, the Experiements work is implemented and being used regularly. There
are several things that need to be implemented in it to make it the best
that it can be, but it works.
>
> Also, I wanted to point out that your support for triggers seems to dock
> pretty directly to something that is an inherent part of the MetaABM / AMF
> design. We have a rich construct for Actions, in fact, my view of an ABM
> is as a set of Queries (in agent, space and time) and Transformations
> (i.e. paramter settings, agent movements, graph connection creation). Of
> course, a binary predicate is just a disembodied query... An aspect of
> this that I think is really unique is that these "Selections" (think SQL
> queries) can be chained, so for example using just queries and
> transofmations you can do things like have all agents withi some
> chacterisitcs search for a related agent within n spaces, move toward that
> agent, create a network connection with it, etc.. All models in the
> MetaABM system are deifned this way and I'm pretty certain that all
> cannonical agent-based models can be. All of the code gen and editors for
> this stuff works and is pretty mature.
>
> -Miles
>
The STEM triggers implementation is still in need of some work. It doesn't
support fully general predicates yet, but it still has proved useful.
Dan
|
|
| | |
| Re: AMP Proposal, Equation / Agent approach [message #2111 is a reply to message #2056] |
Fri, 10 April 2009 21:32 |
Miles Parker
Messages: 1341
Registered: July 2009 |
Senior Member |
|
|
On 2009-04-10 02:28:07 -0700, "Werner Keil" <werner.keil@gmx.net> said:
> Miles,
>
> Thanks a lot for your proactive approach.
> Not all new projects or proposing parties manage to do that.
Well thanks. I've begun to feel like my enthusiasm might be getting a
bit annoying, frankly.
> I noticed a rather large number of healthcare related experts in your proposal.
> Which itself does sound a little more generic, so your project may add
> some building block to the idea I discussed with both STEM and some of
> the Eclipse Board members I know.
Yes, that's true and it is *very* generic. One of the really
interesting aspects of Biomed right now is that people seem to finally
be waking up to the importance of multi-scale / cross-scale issues. NIH
for example seems to have a real interest in multi-scale, but so far we
haven't had much succcess in getting support for the software /
technical side of things. That can be a real problem wtih publicaly
fuded research unfortunatly.. :(
With ABM I've worked with everyone from archelogists to economists to..
well can't say any zoologists that I know about but I guess eclogist is
pretty close... That's one reason that I'd really like to get STEM
folks contributing there ideas and experience even if not code. In
return I'll try to feedback into STEM whenever that makes sense. My
personal overall vision here is a little more expansive ;) -- think
general science platform -- but I think we have enough to chew on for
now, and I think taking a broad agent view really cover a lot of the
ground that folks would be interested in. But I hope that explains my
mention of experiment management etc.. There is a potentially a big
data component here as well -- which may or may not be in scope for
either but dynamic data certainly is part of AMP scope and there can at
least be some incubation of ideas there.
> To help support any kind of economic early warning system or analysis
> of the current (and earlier, data and similarities here exists up to
> 200 years back!!) economic condition.
One caveat I'd always like to give here is regarding explanation vs.
prediction. But the point is that you should be able to see the basic
conditions. Just as we might not know *when* a cholera epidemic will
strike but we do know that there are many local areas where water
systems are breaking down, we may not know at what point the credit
market will collapse, but we know that the basic mechanism of credit
default swaps is not robust to local liquidity shocks and that those
shocks are bound to happen.
So those are the things that one wants to identify to policy makers and
at some point hope they listen. Now I'm hearing that they are in some
cases, but its all been post hoc unfortunatly. Hopefully the same won't
be true with say avian flu.
>
> Proposing some of that to Eclipse Financial (who are in the same phase)
> may help where you consider this a possible synergy, too.
That's a *very* interesting idea, thanks Werner..
|
|
|
| Re: AMP Proposal, Equation / Agent approach [message #2119 is a reply to message #2042] |
Fri, 10 April 2009 21:41 |
Miles Parker
Messages: 1341
Registered: July 2009 |
Senior Member |
|
|
On 2009-04-09 17:35:09 -0700, "Daniel Ford" <daford@almaden.ibm.com> said:
>
> "Miles Parker" <milesparker@gmail.com> wrote in message
> news:griro4$deb$1@build.eclipse.org...Conceptually, this isn't hard to
> imagine, but in practice there would be
> some challenges. The main one would be the synchronization. If we had a
> number of agent models feeding into seperate nodes in a STEM simulation, the
> simulation cycle time would be that of the slowest agent simulation. Right
> now, for the entire United States at the county level (3000+ graph nodes),
> the best cycle time is just a bit less than 500 milliseconds. I suspect
> that the Agent based simulations will be a bit longer and if there are
> multiple instances running in parallel they all need to be managed; that
> would be a challenge.
Right, but that's largely a preformance issue, not architectural..?
Your suspcion would be right of course, but I'm betting ;) that the
potential for High Performance Model Driven approaches is huge. This
whole thing is also embarrasignly parallel, so you could imagine a
cluster node for each model node. Actually, you don't have to imagine
it because it has been done one-off, but it would be ncie to have a
solid public architecture for it. Of course, I don't mean to diminish
latency and synchronization issues and so yes it would probably be a
bit ambitious for a demo. It would be a very interesting proof of
concept / pilot effort though.
>
> STEM can currently support three kinds of geographic visulatizatoin, it's on
> internal view, Google Earth and, more a proof of implemenation, NASA
> Whirlwinds.
Great.
> Yes, the Experiements work is implemented and being used regularly. There
> are several things that need to be implemented in it to make it the best
> that it can be, but it works.
OK, I'll check it out. Any particular entry point that would make the
most sense to look at?
> The STEM triggers implementation is still in need of some work. It doesn't
> support fully general predicates yet, but it still has proved useful.
Take a look at the MetaABM nee AMF meta-model then and see if it might
be a fit. I hadn't thought of seperating it into a completely seperate
..ecore model, but it doesn't ahve any deep dependencies on the rest of
the ABM stuff. I'm just in review on my Docuware and that goes into a
bit of depth on it. I'll update you when that gets posted.
|
|
| |
| Re: AMP Proposal, Equation / Agent approach [message #2236 is a reply to message #2206] |
Tue, 14 April 2009 21:01 |
Miles Parker
Messages: 1341
Registered: July 2009 |
Senior Member |
|
|
On 2009-04-14 05:32:04 -0700, "Daniel Ford" <daford@almaden.ibm.com> said:
> Shifting the content of the thread slightly...
>
> What would it take to interface to AMP?
>
> STEM, with appropriate new code, could issue queries to external data
> sources. We demonstrated this a few years ago by interfacing with a
Cool.
> real-time labratory information system so that we could "inject" labratory
> confirmed cases of Influenza into a simulation running at "wall clock"
> rates. To interface to an Agent based system, I would expect we would do
Neat! Jus thad a thought that it might be interesting to do something
similar with google flu trends..
> something similiar by making some connection across the network to a running
> system and then interpreting the results in the context of the state of a
> disease for a particular city or region that makes up a small part of the
> larger graph being maintained by STEM. Of course, there would need to be
> some input to the Agent simulation, otherwise it's just running in isolation
> and not participating in the bigger picture.
Yeah, that's exactly the approach that I think works..pont bing that
you could mix and match and even use different nodes for
cross-validation of models and real-time data.
> Any thoughts on how one might interface with AMP?
I would think that given we're both on Eclipse platofrm and we want to
exemplify that that an Eclipse specific approach would be most
compelling. I see 3, in addition to the more generic query mechanism
you mention above..
1. Pure API dependencies.
a. Some kind of synchornization / message passing
b. STEM running as actual AMP AXF (exectuon framework) participant,
i.e. AXF controls ABM and STEM steps. The idea of AXF is that it really
would be as domain / methodlogy netural as possible. The only basic
contract as I have planned now (and based on tested current approach)
is that the model has to be able to receive requests to iterate,
iterate, send an "iterated" event, and then wait for the next request
to iterate.
c. Other way around, STEM driven and sending step requests to ABM.
2. Similar to former, but with extension point / adapter approach.
(That coule be nice because it might demonstrate a non-dependency
driven, i.e. truly pluggable integration.)
3. Some kind of direct EMF meta-model reference mediated approach.
Note that because all generated code (say for Escape) is extensible /
adaptable, we could insert arbitrary hooks into generated ABM code as
well.
> Is there an existing
> disease simulation available for AMP?
Yes and no. AMP is really new software though the exemplar system
"Escape" that is planned to be part of the project is API compatible
with Ascape, an ABM platform that has a ten year history. I and others
have certainly done disease simulation within Ascape including
Incubation, Tranismission, Pathogenecity, etc.. on spatial landscapes,
but unfortunatly the work I did was propreitary. If we get into
disussion I might be able to see if collegeaus would be interested in
participating and providing their code. In practice I could work up a
reasonably credible notional model in fairly short order but that would
have to fit in with other priorites and projects naturally.
|
|
|
| Re: AMP Proposal [message #560660 is a reply to message #1958] |
Wed, 08 April 2009 13:29 |
Daniel Ford
Messages: 148
Registered: July 2009
Location: New York |
Senior Member |
|
|
I read the proposal, looks cool. We should definitely look for synergy.
The STEM geographic data sets are pretty extensive and have gone through the
all important and challenging Ecliopse IP CQ process. There's also scope
for interaction as it would be possible to create a kind of hybrid model in
which particular labels in the graph maintained by STEM were updated by
particular Agent based models.
--
Daniel Ford
IBM Almaden Research Center
San Jose, CA
"Miles Parker" <milesparker@gmail.com> wrote in message
news:gre2f2$j6c$1@build.eclipse.org...
> Hi all,
>
> Just waiting for this newsgroup to startup to introduce the STEM team to
> the Agent Modeling Platform. There seems to be a lot of potential for
> collaboration here -- I'm guessing that this is the only other Eclispe
> project that has familiarity with complexity science and modeling (as in
> modeling and simulation, not model driven development) issues. Plus, AMP
> has space and time too. :)
>
> http://eclipse.org/proposals/amp
>
> I'm particularly interested that you seem to have done a lot of work
> supporting GIS -- that is an area that AMP will eventually need direct
> support for but that isn't in project plans. I'd also like to take a look
> in depth at your neta-models to see where there might be connections.
> Another large area directly related to STEM is exploration of automated
> extraction of executable models from ontologies and biomed data-sources.
> (Feel free to send me an email offline as well.)
>
> On the AMP side, we've got a very general high-level meta-model
> representation for agent behaviors that I'd love to get your thoughts on.
> I've also done a lot of work on GUI issues, i.e. generation and runtime
> support, UI, EMF.Edit, etc.. so perhaps there are some pieces that you
> will find useful there. The AMP execution framework is planned to provide
> a lot of model execution time support, including views, GEF EditParts,
> etc., and significant parts of this are working now.
>
> best regards and congratulations,
>
> Miles
>
> Miles Parker
> Metascape, LLC
> http://miles@metascapeabm.com
> http://milesparker.blogspot.com
>
|
|
|
| Re: AMP Proposal, Equation / Agent approach [message #560669 is a reply to message #2013] |
Wed, 08 April 2009 18:51 |
Miles Parker
Messages: 1341
Registered: July 2009 |
Senior Member |
|
|
On 2009-04-08 06:29:16 -0700, "Daniel Ford" <daford@almaden.ibm.com> said:
> I read the proposal, looks cool. We should definitely look for synergy.
> The STEM geographic data sets are pretty extensive and have gone through the
> all important and challenging Ecliopse IP CQ process. There's also scope
> for interaction as it would be possible to create a kind of hybrid model in
> which particular labels in the graph maintained by STEM were updated by
> particular Agent based models.
Thanks Daniel!
Yes, the reason I brought that up is because I wanted to understand the
level of granularity as that would give a sense of where hybrid
interactions might work. Integrating and / or deriving agent and
equation model representationshas has been something I've been thinking
about for a long time. There are some really cool potentials here --
I've had a similar discuscion with Lyle Wallis ("interested party")
about integraton with dynamic systems models. Its funny, because when
you turn to abstract representation there is a strong sense in which
the distinction if more methodological than technological, esp. WRT
granularity. This is proably way too much to get into in depth in this
forum, but I do have some exposure ;) to and collegeuas in this area,
and could imagine a very nice demo model where say individual node
state comes from a contained agent-based representation. Something that
did a micro representation of contact, transmission, incubation,
etc..that fed into a macro model of population centers. Let me know if
you want to get into this further offline.
I'm anxious to get something available from the toolset soon so that
people can see UI execution aspects. Perhaps I'll try to get some
screencasts done. And the geostuff is really important WRT to
visualization and integration -- to have all the world geo sets
provided by the toolset directly would itself be extremely cool.
One other area that is ripe: I am finally in the midst of getting into
work with parameterization and model run (batch) control. I've done a
bit of support for this already; allowing text edited files to drive
model execution and validation / verification testing. But my intention
all along has been to have this be defined through an Ecore model. (See
the score paper referrnced in proposal.) The MetaABM nee AMF
representation already has a quite rich generic representation for
model attributes / parameters. There is no reason in the world that
this has to be taken as an ABM only construct; in fact generalization
to equation and agent representation was kind of a background design
goal. So what jumped out at me was your 0.3.0 description of support
for experiments over scenarios. If you've done this work already then
perhaps there in some obvious way to generalize / merge that
functionality.
Also, I wanted to point out that your support for triggers seems to
dock pretty directly to something that is an inherent part of the
MetaABM / AMF design. We have a rich construct for Actions, in fact, my
view of an ABM is as a set of Queries (in agent, space and time) and
Transformations (i.e. paramter settings, agent movements, graph
connection creation). Of course, a binary predicate is just a
disembodied query... An aspect of this that I think is really unique is
that these "Selections" (think SQL queries) can be chained, so for
example using just queries and transofmations you can do things like
have all agents withi some chacterisitcs search for a related agent
within n spaces, move toward that agent, create a network connection
with it, etc.. All models in the MetaABM system are deifned this way
and I'm pretty certain that all cannonical agent-based models can be.
All of the code gen and editors for this stuff works and is pretty
mature.
-Miles
|
|
|
| Re: AMP Proposal, Equation / Agent approach [message #560679 is a reply to message #2027] |
Fri, 10 April 2009 00:35 |
Daniel Ford
Messages: 148
Registered: July 2009
Location: New York |
Senior Member |
|
|
"Miles Parker" <milesparker@gmail.com> wrote in message
news:griro4$deb$1@build.eclipse.org...
> On 2009-04-08 06:29:16 -0700, "Daniel Ford" <daford@almaden.ibm.com> said:
>
>> I read the proposal, looks cool. We should definitely look for synergy.
>> The STEM geographic data sets are pretty extensive and have gone through
>> the
>> all important and challenging Ecliopse IP CQ process. There's also scope
>> for interaction as it would be possible to create a kind of hybrid model
>> in
>> which particular labels in the graph maintained by STEM were updated by
>> particular Agent based models.
>
> Thanks Daniel!
>
> Yes, the reason I brought that up is because I wanted to understand the
> level of granularity as that would give a sense of where hybrid
> interactions might work. Integrating and / or deriving agent and equation
> model representationshas has been something I've been thinking about for a
> long time. There are some really cool potentials here --
> I've had a similar discuscion with Lyle Wallis ("interested party") about
> integraton with dynamic systems models. Its funny, because when you turn
> to abstract representation there is a strong sense in which the
> distinction if more methodological than technological, esp. WRT
> granularity. This is proably way too much to get into in depth in this
> forum, but I do have some exposure ;) to and collegeuas in this area, and
> could imagine a very nice demo model where say individual node state comes
> from a contained agent-based representation. Something that did a micro
> representation of contact, transmission, incubation, etc..that fed into a
> macro model of population centers. Let me know if you want to get into
> this further offline.
Conceptually, this isn't hard to imagine, but in practice there would be
some challenges. The main one would be the synchronization. If we had a
number of agent models feeding into seperate nodes in a STEM simulation, the
simulation cycle time would be that of the slowest agent simulation. Right
now, for the entire United States at the county level (3000+ graph nodes),
the best cycle time is just a bit less than 500 milliseconds. I suspect
that the Agent based simulations will be a bit longer and if there are
multiple instances running in parallel they all need to be managed; that
would be a challenge.
>
> I'm anxious to get something available from the toolset soon so that
> people can see UI execution aspects. Perhaps I'll try to get some
> screencasts done. And the geostuff is really important WRT to
> visualization and integration -- to have all the world geo sets provided
> by the toolset directly would itself be extremely cool.
STEM can currently support three kinds of geographic visulatizatoin, it's on
internal view, Google Earth and, more a proof of implemenation, NASA
Whirlwinds.
>
> One other area that is ripe: I am finally in the midst of getting into
> work with parameterization and model run (batch) control. I've done a bit
> of support for this already; allowing text edited files to drive model
> execution and validation / verification testing. But my intention all
> along has been to have this be defined through an Ecore model. (See the
> score paper referrnced in proposal.) The MetaABM nee AMF representation
> already has a quite rich generic representation for model attributes /
> parameters. There is no reason in the world that this has to be taken as
> an ABM only construct; in fact generalization to equation and agent
> representation was kind of a background design goal. So what jumped out at
> me was your 0.3.0 description of support for experiments over scenarios.
> If you've done this work already then perhaps there in some obvious way to
> generalize / merge that functionality.
Yes, the Experiements work is implemented and being used regularly. There
are several things that need to be implemented in it to make it the best
that it can be, but it works.
>
> Also, I wanted to point out that your support for triggers seems to dock
> pretty directly to something that is an inherent part of the MetaABM / AMF
> design. We have a rich construct for Actions, in fact, my view of an ABM
> is as a set of Queries (in agent, space and time) and Transformations
> (i.e. paramter settings, agent movements, graph connection creation). Of
> course, a binary predicate is just a disembodied query... An aspect of
> this that I think is really unique is that these "Selections" (think SQL
> queries) can be chained, so for example using just queries and
> transofmations you can do things like have all agents withi some
> chacterisitcs search for a related agent within n spaces, move toward that
> agent, create a network connection with it, etc.. All models in the
> MetaABM system are deifned this way and I'm pretty certain that all
> cannonical agent-based models can be. All of the code gen and editors for
> this stuff works and is pretty mature.
>
> -Miles
>
The STEM triggers implementation is still in need of some work. It doesn't
support fully general predicates yet, but it still has proved useful.
Dan
|
|
| | |
| Re: AMP Proposal, Equation / Agent approach [message #560719 is a reply to message #2056] |
Fri, 10 April 2009 21:32 |
Miles Parker
Messages: 1341
Registered: July 2009 |
Senior Member |
|
|
On 2009-04-10 02:28:07 -0700, "Werner Keil" <werner.keil@gmx.net> said:
> Miles,
>
> Thanks a lot for your proactive approach.
> Not all new projects or proposing parties manage to do that.
Well thanks. I've begun to feel like my enthusiasm might be getting a
bit annoying, frankly.
> I noticed a rather large number of healthcare related experts in your proposal.
> Which itself does sound a little more generic, so your project may add
> some building block to the idea I discussed with both STEM and some of
> the Eclipse Board members I know.
Yes, that's true and it is *very* generic. One of the really
interesting aspects of Biomed right now is that people seem to finally
be waking up to the importance of multi-scale / cross-scale issues. NIH
for example seems to have a real interest in multi-scale, but so far we
haven't had much succcess in getting support for the software /
technical side of things. That can be a real problem wtih publicaly
fuded research unfortunatly.. :(
With ABM I've worked with everyone from archelogists to economists to..
well can't say any zoologists that I know about but I guess eclogist is
pretty close... That's one reason that I'd really like to get STEM
folks contributing there ideas and experience even if not code. In
return I'll try to feedback into STEM whenever that makes sense. My
personal overall vision here is a little more expansive ;) -- think
general science platform -- but I think we have enough to chew on for
now, and I think taking a broad agent view really cover a lot of the
ground that folks would be interested in. But I hope that explains my
mention of experiment management etc.. There is a potentially a big
data component here as well -- which may or may not be in scope for
either but dynamic data certainly is part of AMP scope and there can at
least be some incubation of ideas there.
> To help support any kind of economic early warning system or analysis
> of the current (and earlier, data and similarities here exists up to
> 200 years back!!) economic condition.
One caveat I'd always like to give here is regarding explanation vs.
prediction. But the point is that you should be able to see the basic
conditions. Just as we might not know *when* a cholera epidemic will
strike but we do know that there are many local areas where water
systems are breaking down, we may not know at what point the credit
market will collapse, but we know that the basic mechanism of credit
default swaps is not robust to local liquidity shocks and that those
shocks are bound to happen.
So those are the things that one wants to identify to policy makers and
at some point hope they listen. Now I'm hearing that they are in some
cases, but its all been post hoc unfortunatly. Hopefully the same won't
be true with say avian flu.
>
> Proposing some of that to Eclipse Financial (who are in the same phase)
> may help where you consider this a possible synergy, too.
That's a *very* interesting idea, thanks Werner..
|
|
|
| Re: AMP Proposal, Equation / Agent approach [message #560728 is a reply to message #2042] |
Fri, 10 April 2009 21:41 |
Miles Parker
Messages: 1341
Registered: July 2009 |
Senior Member |
|
|
On 2009-04-09 17:35:09 -0700, "Daniel Ford" <daford@almaden.ibm.com> said:
>
> "Miles Parker" <milesparker@gmail.com> wrote in message
> news:griro4$deb$1@build.eclipse.org...Conceptually, this isn't hard to
> imagine, but in practice there would be
> some challenges. The main one would be the synchronization. If we had a
> number of agent models feeding into seperate nodes in a STEM simulation, the
> simulation cycle time would be that of the slowest agent simulation. Right
> now, for the entire United States at the county level (3000+ graph nodes),
> the best cycle time is just a bit less than 500 milliseconds. I suspect
> that the Agent based simulations will be a bit longer and if there are
> multiple instances running in parallel they all need to be managed; that
> would be a challenge.
Right, but that's largely a preformance issue, not architectural..?
Your suspcion would be right of course, but I'm betting ;) that the
potential for High Performance Model Driven approaches is huge. This
whole thing is also embarrasignly parallel, so you could imagine a
cluster node for each model node. Actually, you don't have to imagine
it because it has been done one-off, but it would be ncie to have a
solid public architecture for it. Of course, I don't mean to diminish
latency and synchronization issues and so yes it would probably be a
bit ambitious for a demo. It would be a very interesting proof of
concept / pilot effort though.
>
> STEM can currently support three kinds of geographic visulatizatoin, it's on
> internal view, Google Earth and, more a proof of implemenation, NASA
> Whirlwinds.
Great.
> Yes, the Experiements work is implemented and being used regularly. There
> are several things that need to be implemented in it to make it the best
> that it can be, but it works.
OK, I'll check it out. Any particular entry point that would make the
most sense to look at?
> The STEM triggers implementation is still in need of some work. It doesn't
> support fully general predicates yet, but it still has proved useful.
Take a look at the MetaABM nee AMF meta-model then and see if it might
be a fit. I hadn't thought of seperating it into a completely seperate
..ecore model, but it doesn't ahve any deep dependencies on the rest of
the ABM stuff. I'm just in review on my Docuware and that goes into a
bit of depth on it. I'll update you when that gets posted.
|
|
| |
| Re: AMP Proposal, Equation / Agent approach [message #560768 is a reply to message #2206] |
Tue, 14 April 2009 21:01 |
Miles Parker
Messages: 1341
Registered: July 2009 |
Senior Member |
|
|
On 2009-04-14 05:32:04 -0700, "Daniel Ford" <daford@almaden.ibm.com> said:
> Shifting the content of the thread slightly...
>
> What would it take to interface to AMP?
>
> STEM, with appropriate new code, could issue queries to external data
> sources. We demonstrated this a few years ago by interfacing with a
Cool.
> real-time labratory information system so that we could "inject" labratory
> confirmed cases of Influenza into a simulation running at "wall clock"
> rates. To interface to an Agent based system, I would expect we would do
Neat! Jus thad a thought that it might be interesting to do something
similar with google flu trends..
> something similiar by making some connection across the network to a running
> system and then interpreting the results in the context of the state of a
> disease for a particular city or region that makes up a small part of the
> larger graph being maintained by STEM. Of course, there would need to be
> some input to the Agent simulation, otherwise it's just running in isolation
> and not participating in the bigger picture.
Yeah, that's exactly the approach that I think works..pont bing that
you could mix and match and even use different nodes for
cross-validation of models and real-time data.
> Any thoughts on how one might interface with AMP?
I would think that given we're both on Eclipse platofrm and we want to
exemplify that that an Eclipse specific approach would be most
compelling. I see 3, in addition to the more generic query mechanism
you mention above..
1. Pure API dependencies.
a. Some kind of synchornization / message passing
b. STEM running as actual AMP AXF (exectuon framework) participant,
i.e. AXF controls ABM and STEM steps. The idea of AXF is that it really
would be as domain / methodlogy netural as possible. The only basic
contract as I have planned now (and based on tested current approach)
is that the model has to be able to receive requests to iterate,
iterate, send an "iterated" event, and then wait for the next request
to iterate.
c. Other way around, STEM driven and sending step requests to ABM.
2. Similar to former, but with extension point / adapter approach.
(That coule be nice because it might demonstrate a non-dependency
driven, i.e. truly pluggable integration.)
3. Some kind of direct EMF meta-model reference mediated approach.
Note that because all generated code (say for Escape) is extensible /
adaptable, we could insert arbitrary hooks into generated ABM code as
well.
> Is there an existing
> disease simulation available for AMP?
Yes and no. AMP is really new software though the exemplar system
"Escape" that is planned to be part of the project is API compatible
with Ascape, an ABM platform that has a ten year history. I and others
have certainly done disease simulation within Ascape including
Incubation, Tranismission, Pathogenecity, etc.. on spatial landscapes,
but unfortunatly the work I did was propreitary. If we get into
disussion I might be able to see if collegeaus would be interested in
participating and providing their code. In practice I could work up a
reasonably credible notional model in fairly short order but that would
have to fit in with other priorites and projects naturally.
|
|
|
Goto Forum:
Current Time: Mon Sep 23 01:22:05 GMT 2024
Powered by FUDForum. Page generated in 0.06374 seconds |
] 
Search
Help
Register
Login
Home
