Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-1
Model ModelWorkshop
- relating simulation models
At EHESS, GREQUAM/CNRS
Marseille, 2003
Welcome to the…
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-2
Organisation
Juliette Rouchier
durandal.cnrs-mrs.fr/GREQAM/cv/rouchier.htm
David Hales
www.davidhales.com
Bruce Edmonds
bruce.edmonds.name
EHESS, GREQUAM/CNRS
durandal.cnrs-mrs.fr/ehess/ehess.html
Centre for Policy Modellingcfpm.org
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-3
Relating Models- an overview
Bruce Edmonds(including material from David Hales
and Juliette Rouchier)
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-4
Outline
1. Kinds of model, in particular comparing equation-based modelling and individual/agent-based simulation
2. Some basic ways in which models may be related or compared
3. Some uses for relating or comparing models
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-5
Part 1: Overview and Comparison of Equation-based Modelling and
Agent-based Simulation
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-6
Equation-based modellingModel Target Equation-based Model
Actual Outcomes
AggregatedActual Outcomes
AggregatedModel Outcomes
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-7
Properties of (equation-based) Mathematical Models
• Long tradition, many techniques/results
• Central use of numbers and proof
• In simple cases can derive closed-form (i.e. general) conclusions
• Essentially about states– Atemporal (where time occurs it is reified)– Inference can works in may ways
• Assumptions necessary to represent world
• ‘Art’ of approximation and application
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-8
Individual-based simulation
Model Target Agent-based Model
Actual Outcomes Model Outcomes
AggregatedActual Outcomes
AggregatedModel Outcomes
Agent-based
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-9
Properties of (agent/individual-based) Simulations
• Short tradition, fewer techniques/results• Central use of algorithms and computation• Difficult to derive general conclusions
– More like an experiment than an inference
• Essentially about process– Temporal directionality– Process of unfolding observable
• More representational in practice– In time and in composition
• More suggestive of interpretation
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-10
Some Uses of Mathematics
• To predict what is currently unknown• To explain what is already known• To derive/infer conclusions from axioms• To represent observed phenomena• To explore unobserved possibilities• To make an idea unambiguous• To compress a representation• To show one model is a special case of
another• To construct a formal framework/language
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-11
Some Uses of Simulation
• To predict what is currently unknown• To explain what is already known• To calculate outcomes from initial set-up• To represent observed phenomena• To explore unobserved possibilities• To make an idea unambiguous• To compress a representation (in theory)• To show one model is a special case of
another (in theory)• To construct a formal framework/language
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-12
Part 2: Some Ways in which Models may be
Compared or Related
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-13
A diagram for a simulation model
Code
Agents
Setting
Outcomes from one run
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-14
As a Summary/Abstraction of Another Model’s Results
• Modelling model results• More abstract model can be of any kind (including
simulations and equation-based models)• Can be a tactic to help understand/analyse complex
models
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-15
As a Generalisation/Specialisation of Another Model
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-16
As a Controlled Experiment/Comparison
?
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-17
Summary of Some Basic Kinds of Model Relation/Comparison
• Abstraction/approximation of another model
• Super/subset of another model
• As a Controlled Comparison
• The code of one model is a component of the code of another
• Others … ?
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-18
Part 3: Some Uses for Relating Models
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-19
To check models are equivalent
• Examination of the code suggests a hypothesis that the simulations are equivalent but…
• …this can only be disconfirmed by experiment• Repeated failure to disconfirm hypothesis can lead one to
rely on it• Reveal hidden assumptions – understand the limits of our
creations!
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-20
As a tool for staging abstraction
Observations of the
phenomena
Descriptive simulation
Equation-based models
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-21
Comparison to Inform Generalisation
comparison
GeneralisedModel
Model to Model Workshop, EHESS, Grequam/CNRS, Marseille 2003, http://cfpm.org/m2m slide-22
Summary of Some Uses of Relating Models• Check if models are the same
• To reveal assumptions
• As a tool for staging abstraction
• To inform generalisation
• Communicate and compare complex phenomena between researchers
• And … ?