research challenges faced the agent-based modelling framework required the following features:...

12
Research challenges faced The Agent-based modelling framework required the following features: Ability to run many millions of complex agents Should run on any supercomputer or desktop (Linux, Mac, Windows) Allows economists to design models with no knowledge of programming Strong testing techniques to assure quality – most large scientific software has serious bugs [Hatton 1997] Ability to combine different models sharing agents in a coherent and correct way

Upload: cody-nichols

Post on 26-Dec-2015

221 views

Category:

Documents


7 download

TRANSCRIPT

Page 1: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

Research challenges faced• The Agent-based modelling framework required

the following features:– Ability to run many millions of complex agents– Should run on any supercomputer or desktop (Linux,

Mac, Windows)– Allows economists to design models with no

knowledge of programming– Strong testing techniques to assure quality – most

large scientific software has serious bugs [Hatton 1997]

– Ability to combine different models sharing agents in a coherent and correct way

Page 2: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

Basic philosophy of FLAME• Agents defined as communicating X-machines

– X-machine concept introduced by Eilenberg [1974] and then ignored until recently

• General computational model• Intuitive and easy to use in many different

applications – Economics, biology, management, sociology

• Messages sent to message boards• Agents are distributed across processors using

new techniques• Efficient implementation utilises the available

hardware – serial or parallel

Page 3: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

What agents do?

• They have a memory – it contains their identity, location, other information such as their assets, employment state etc.

• They have a state which determines what they can do next

• They can read messages sent to them• They can write messages to other agents• They can carry out other operations if these are

permitted – accept job offer, buy shares, apply for credit etc.

Page 4: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

The X-machine

n.xmlstate, positional informationetc. Input to external solvers.Input to post-processing.

messages

functions - mathematical - logical

internal functions orexternal e.g. COPASI

0.xml

initial conditions

Page 5: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

FLAME Block Diagram

X parser files

X parser files

Xparser.exe

Xparser.exe

Model.xmlModel.xml

Functions.cFunctions.c

1-NXml files

1-NXml files

Main.exeMain.exe

0.xml0.xml

make

LibmboardLibmboard

Your filesYour files Xparser files

Xparser files

Page 6: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

How do you define agents?

• Use XMML – e.g.

etc.

Page 7: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

Firm in labour market – state table

Page 8: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

A firm’s functions• Firm send vacancies. If additional workers are needed the

firm sends vacancies messages– especially the different wage offers for the different

general skill groups.• Firm send redundancies. If the firm wants to decrease the

workforce it sends redundancies.• Firm idle. Firm does nothing.• Firm reads job applications sends job offer or rejection.• Firm reads the application, ranks the applicants according to

their general and specific skills and sends as many job offers to the first ranked applicants as the firm has vacancies to fill. The other applicants are refused.

• Firm read job responses. The firm reads the responses to their job offers and updates the number of employees and the number of vacancies.

Page 9: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

Firm messages

Page 10: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should
Page 11: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

Parallel implementation

Parallel agents grouped on parallel nodes.

Messages synchronised

Message board library allows both serial and parallel versions to work

Implementation details hiddenfrom modellers

System automatically manages the simulation

Page 12: Research challenges faced The Agent-based modelling framework required the following features: –Ability to run many millions of complex agents –Should

C. Greenhough, D. Worth