Skip to main content
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [stem-dev] Interfacing STEM with external workflows RE: Committing a Disease Model variant

My understanding is that there is a important external program that determines the parameters of the STEM scenario in perhaps ways we cannot determine beforehand, so like option 2 below.

Without any changes to the current code, this could be done in a shell script

1. Call external program
2. Read and parse the output, overwrite the $$$.standard (disease model or perhaps an infector) file under the decorators folder with new parameters.
3. Call 'STEM.exe -headless -uri <your scenario> -log"
4. Repeat from step 1 as long as needed.

Regards,
/ Stefan


Stefan Edlund
Public Health and Computer Science Research
IBM Almaden Research Center
(408) 927-1766   edlund@xxxxxxxxxxxxxxx


Inactive hide details for James Kaufman ---11/22/2011 09:52:23 AM---Jan, Let's take this discussion to stem-dev (added above).James Kaufman ---11/22/2011 09:52:23 AM---Jan, Let's take this discussion to stem-dev (added above).


    From:

James Kaufman <kaufman@xxxxxxxxxxxxxxx>

    To:

"Wigger Jan-Frederik" <Jan-Frederik.Wigger@xxxxxxxxxxx>, stem-dev@xxxxxxxxxxx,

    Date:

11/22/2011 09:52 AM

    Subject:

[stem-dev] Interfacing STEM with external workflows RE: Committing a Disease Model variant

    Sent by:

stem-dev-bounces@xxxxxxxxxxx



Jan, 
Let's take this discussion to stem-dev (added above). 

I'm still not fully understanding your goal. Are any of the following on target? 

1) If your goal is to use an external program to specify a list of parameters to use in
kicking off a series of STEM runs, then STEM has a mechanism you can take advantage of. Please read on the wiki about 
STEM "Experiments". http://wiki.eclipse.org/Batch_Mode_%28Running_Experiments%29 
Instead of using the STEM wizard or GUI, all you would need to do is to have your external process or workflow 
modify the $$$.experiment file. Then run it. We could probably set up a way to kick off a headless run of an Experiment batch 
just like we now do for a single scenario. 

2) If your goal is to modify the parameters of a single active simulation before it runs, then I think it makes more sense to modify the 
the disease .decorator file (in the decorators folder). This is preferable to making an external call at runtime (on startup) because if you take 
that option the log files will be wrong. If you modify the .decorator file, then STEM will correctly log the parameters actually run and keep those with the
results (same for option 1 above). 

3) The other case I can image is that you want to use a statistical package like R to change the disease parameters until your runs find a minimum or until a model 
fits some external data. Here again I think we want to try to take advantage of another existing feature, namely "automated experiments". 
Please see: http://wiki.eclipse.org/Running_an_Automated_Experiment 
This feature uses STEMS Nelder-Mead simplex algoritm to do a fit to data. You could have your workflow modify the $$$.automaticexperiment file before kicking it off. 

4) If none of the above I would definitely like to have a phone call so Stefan and Matt can join this discussion. Please keep in mind that if we create a new disease model 
that makes and external call in the "updatelabels()" method, this method is called at every iteration and at every node in the canonical graph.

Thanks in advance for your patience. I really want to make sure we understand the use case so we start down the right path.... 

Sincerely,
Jamie

IBM Almaden Research Center, 650 Harry Rd.
San Jose, CA 95120-6099
email: kaufman@xxxxxxxxxxxxxxx
phone: (408) 927-2477  (tie 457-2477)
    From: 
    "Wigger Jan-Frederik" <Jan-Frederik.Wigger@xxxxxxxxxxx> 
    To: 
    "Kaufman James" <kaufman@xxxxxxxxxxxxxxx>,
    Cc: 
    "Edlund Stefan" <edlund@xxxxxxxxxxxxxxx>, "Weiser Armin" <Armin.Weiser@xxxxxxxxxxx>, "Filter Matthias" <Matthias.Filter@xxxxxxxxxxx> 
    Date: 
    11/22/2011 06:20 AM 
    Subject: 
    Antw: Re: Committing a Disease Model variant




Hey Jamie,

thank you very much for your quick response.
I think it is something similar to (1) what we need, but much easier.
We want to call something like R-scripts or other external programs,
that return the results of their calculations in a standard output
format.
Something like this:
Runtime rt = Runtime.getRuntime();
Process pr = rt.exec("c:\\..\\R.exe --vanilla --slave  <
C:\\..\\own_model.R ");

This is then read by the call within STEM, like:
BufferedReader input = new BufferedReader(new
InputStreamReader(pr.getInputStream()));
...

These values are then used to set the parameters of the disease model.

So all of this is fairly low-level stuff, but supposed to be the first
step in the development of a communication with external models that
calculate parameters of a simulation.

If there is another way of handling this or covered by Stefans
approach, any help with this is as always highly appreciated.

Thanks!
Jan


>>> James Kaufman <kaufman@xxxxxxxxxxxxxxx> 11/21/2011 5:40  >>>
Hi Jan,
We already have something called an external data source disease model.

Please take a look at this code and see if it is a model that might
work.
It is not multipopulation but it should be possible to generalize it to

multipopulation. It basically reads files and "plays them back".

Is your goal primarily to
1) Use STEM to initialize and kick off a separate simulation
or
2) Read and use results from and external simulation inside STEM.

The externaldatasourcedisease model handles 2
Stefan has protoytped a new feature called "compile" for case (1) but
we
really need to think carefully about the API for that before committing

it. No call this week because of thanksgiving but let's take this
discussion forward on stem-dev.

Best Regards,
Jamie

IBM Almaden Research Center, 650 Harry Rd.
San Jose, CA 95120-6099
email: kaufman@xxxxxxxxxxxxxxx
phone: (408) 927-2477  (tie 457-2477)





From:
"Wigger Jan-Frederik" <Jan-Frederik.Wigger@xxxxxxxxxxx>
To:
"Edlund Stefan" <edlund@xxxxxxxxxxxxxxx>, "Kaufman James"
<kaufman@xxxxxxxxxxxxxxx>,
Cc:
"Weiser Armin" <Armin.Weiser@xxxxxxxxxxx>, "Filter Matthias"
<Matthias.Filter@xxxxxxxxxxx>
Date:
11/21/2011 04:47 AM
Subject:
Committing a Disease Model variant



Hey Stefan and Jamie,

we've got another question next to the discussion about the
cattle-meat-case..
In our project, we want to realize the possibility to call external
models out of STEM. The easiest way for us that still meets the
project
requirements seems to be the following:
in the new-disease-wizard, there should be a button like "use external
model".
By pressing the button, a new dialog appears where you can give the
path of the external model program.
This external model program reads whatever values have been set in the
disease dialog or ignore the disease parameters if no values have been
set.
The external model returns values for the disease model and sets them.

So, at this level, this is only a function for bonelazy users that
don't want to
- start an external program,
- feed it and
-type in the results into STEM
themselves.
So there is no big intellectual challenge in this. But it is supposed
to show the beginning of this direction that might give a good
impression in the project report.

Are you okay when I create a new disease model like "Multipopulation
SEIR disease model + external" that I commit?
New stuff would only be the button and its function to call an
external
program and the dialog to enter the model path and parameters.

Kind regards
Jan


>>> Stefan Edlund <edlund@xxxxxxxxxxxxxxx> 11/11/2011 10:25  >>>
Hi Jan,

it's really exciting what you are trying to do and we're eager to help
out
in any way we can. Food mediated diseases could be a very important
direction to take STEM, also very timely given the recent high
visibility
outbreaks.

Here's some answers/suggestions to your questions from me in blue and
Jamie
in  green.

 1)            It might be easier to model the scenario with only 2
populations:
    animal and human. It has the advantage that an infected animal
from
    the farm passes through the slaughterhouse and the infection of
    humans could be realized via a mulit-population disease model.
    Problem in this setting is, that infected animals in the area 1
    country nodes would behave like a regular animal population
which
    means that the disease would also spread between the animals in
this
    node which is in contrast to the intended use. Question: is it
    possible to create something like a node specific trigger that
    prevent the transmission of a disease in a specific node? Or
even
    better: is it possible to alter the transmission rates for
specific
    nodes for a disease?

               The interventions framework (
    http://wiki.eclipse.org/Triggering_interventions) is supposed to
be
    able to handle situations like this. The idea is you attach an
    intervention label to a node and the disease model calculation
takes
    into account any control/interventions that are in place.
However,
    currently we model vaccinations and isolations only, and only
for
the
    deterministic SI, SIR and SEIR models (not the multi-population
    disease models). Being able to modify the transmission rate in a
node
    is a great idea, it's essentially saying that social distancing
is
    taking place locally. It's easy to extend STEM and implement it
for
    disease models where it makes sense and if you think it is of
use.

               JK: Rather than using interventions I like the idea of

defining
new
    types of nodes (farms, feeding houses, dairies, etc). With a new
    disease model we could scale the transmission rate based on the
TYPE
    of node.

 2)            Instead of two separate populations for animal and
meat,
we
could
    work with an animal population with “demographic”
subgroups..
    Question: Would this be beneficial?

               I think I see one problem with that. Demographic
subgroups are
really
    for dividing populations into things like men/women. The problem
is
    that the birth/death rate is the same for both subgroups and if
one
    of the subgroups, in your case meat, goes away via migration no
new
    meat is "born". If have a feeling that what you did is correct,
    modeling meat as its own population, and that we need to design
a
new
    type of population model that depends on the availability of
another
    population. Actually, come to think of it this might be useful
for
    modeling parasites too.

               JK. Perhaps 1) Modeling cattle => meat could be a
transition
between
    two compartments in a new type of demographic food (cattle)
    population model. We would use the death rate only to specify
death
    rates for cattle. We could introduce a new rate= "expiration
rate"
    for meat. Dead cattle are never turned into meat (unless we want
to
    model pet food). The transition rate from cattle to meat (with a
    multiplier) would be a mass action term very much like the
infection
    process (SI) term in a disease. The transition rate at which
meat
is
    produced is a function of both the 'S'ource cattle population
    (locally) and the 'I'slaughter house capacity. This could be a
label
    on a new slaughter house type node. There would also be a rate
factor
    beta (production rate) which could be shut off in an
intervention. A
    slaughter house could be a new type of region node with a label
for
    capacity, or it could be a label on any region node. The former
might
    be better because if we want to model e-coli then a slaughter
house
    itself could become infected.
    2) Consumption of meat is like a death rate for the meat
population.
    Again there is a mass action term where meet is consumed by a
product
    of 'S'upply of meat and 'I'human population with a rate term. We
    probably want another rate for expiration of uneaten meat. Doing
it
    this way the rate of consumption (like beta) could change. For
    example if an outbreak occurs people might stop eating the meat
but
    it could get thrown away when it expires. We want to be able to
model
    interventions.
    3) Eventually we should think also about a cattle feed model.
Ideally
    our model for cattle should also support modeling their food.

 3)            Could a Modifier/Trigger set the population A in a
node
to the
value
    of Population B?

               No, we don't have any modifiers that can read the
count
of one
    population (or any other attribute for that matter) and set a
value
    of another population.

      JK We do need to deal with the multiplier between cattle and
meat
    (they don't add up to P). I think in a new demographic
population
    model we could define a scale factor so that
      (A*living cattle)+(B*Meat) = total population or supply. In
this
    case A = 1, B = some multiplier. I'm not sure we would really
use
    total population supply to calculate anything but it could be
used in
    unit testing for death rates = 0.

 4)            Is it possible to have a demographic population model

with in a
node
    that other nodes do not have? For example: California has a
    demographic population model with 2 subpopulations, human1 and
    human2. Is it possible to have a Migration Edge to Nevada that
has a
    Standard population model that only provides a human population?

               No, there is currently no connection between subgroups

and the
    "parent" group that created them. I'm not saying it wouldn't be
    possible to take it into account when processing migration
edges,
but
    it has other side effects I would have to think about. If we
were
to
    do this I think any changes to the human1/human2 population
would
    need to be aggregated up to the 'human' population, something we
    don't do today.

               Correct, no. But we don't need to. IF we made the
transition
between
    cattle and meat a mass action term that depends on the region
being a
    slaughter house, no transition would ever occur outside the
slaughter
    house. Supply of cattle to the slaughter house and distribution
of
    meat out of it would be driven by separate migration edges with
rates
    that can be modified by supply and demand and/or by
interventions.

 As for your exception, I know where it's coming from and it would
need
    some fixing on our side. For now as a workaround, you can set
the
    number of worker threads to 1 (Window->Preferences STEM->Solver
then
    "Number of concurrent worker threads").

 Can you open a defect and attach this scenario? The problem is
really
    how we determine which nodes a thread updates, especially in
this
    case when you have population models that only works on a single
    node.

 Thanks!
 / Stefan



Stefan Edlund
Public Health and Computer Science Research
IBM Almaden Research Center
(408) 927-1766   edlund@xxxxxxxxxxxxxxx





From:       "Wigger Jan-Frederik" <Jan-Frederik.Wigger@xxxxxxxxxxx>



To:         "Edlund Stefan" <edlund@xxxxxxxxxxxxxxx>, "Kaufman
James"
<kaufman@xxxxxxxxxxxxxxx>,


Cc:         "Weiser Armin" <Armin.Weiser@xxxxxxxxxxx>, "Filter
Matthias" <Matthias.Filter@xxxxxxxxxxx>



Date:       11/11/2011 01:23 AM



Subject:    Advice from the sages..








Hey Jamie and Stefan,

It's been quite a while, hope you guys do fine?
As you know, we are working on the problem to model the transmission
of
a disease from a product to human population also considering
production
& trade networks.
We are now wondering how we can best model that with stem with the
least effort possible. Maybe even without touching the code at all..
I created a little toy model that has a farm, a slaughterhouse and
three country nodes. I added some functionalities to the graph editor
to
add edges (and edit their values) and display overlayed edges
correctly,
I will uplaod the update soon...

Next to the model, there is a bunch of text attached where we made an
epic out of what we try to do, if you feel like having a look at the
model and the problem description, any advice to solve the problem
would
be highly appreciated.

Thank you very much and have a nice weekend!

Kind regards

Jan



P.S.
after a while, the simulation throws an error (something with the
disease model), any idea why?
!ENTRY org.eclipse.core.jobs 4 2 2011-11-11 09:50:00.868
!MESSAGE An internal error occurred during: "Finite Difference Worker
1".
!STACK 0
java.util.ConcurrentModificationException
                                at
org.eclipse.emf.common.util.AbstractEList$EIterator.checkModCount
(AbstractEList.java:762)
                                at
org.eclipse.emf.common.util.AbstractEList$EIterator.doNext
(AbstractEList.java:716)
                                at
org.eclipse.emf.common.util.AbstractEList$EIterator.next
(AbstractEList.java:696)
                                at
org.eclipse.stem.diseasemodels.standard.impl.StandardDiseaseModelImpl.applyExternalDeltas
(StandardDiseaseModelImpl.java:327)
                                at
org.eclipse.stem.solvers.fd.impl.FiniteDifferenceImpl.computeDeltasStep
(FiniteDifferenceImpl.java:260)
                                at
org.eclipse.stem.solvers.fd.impl.FdJob.run(FdJob.java:63)
                                at
org.eclipse.core.internal.jobs.Worker.run(Worker.java:54)





Jan-Frederik Wigger
Bundesinstitut für Risikobewertung
Fachgruppe Epidemiologie und Zoonosen
Abteilung Biologische Sicherheit

Federal Institute for Risk Assessment
Unit Epidemiology and Zoonoses
Department Biological Safety
Diedersdorfer Weg 1, 12277 Berlin, Germany
Tel. +49 30 18412-2118
Fax +49 30 18412-2952
www.bfr.bund.de 



[attachment "Bfr_toymodel.zip" deleted by Stefan Edlund/Almaden/IBM]
[attachment "Problemspecification_BfR.rtf" deleted by Stefan
Edlund/Almaden/IBM]





_______________________________________________
stem-dev mailing list
stem-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/stem-dev


GIF image

GIF image

Back to the top