Introduction of abm
Embed Size (px)
Transcript of Introduction of abm
Dr. Yudi Limbar Yasik mail: [email protected]
Agent Based Modeling (ABM) Introduction
o Complex and non linear systems
o Not easy to approach by mathematical or
o Bottom up aproach
o System simulation
o Scenario, prediction
o Ethical problem, Non Parametrik
o Computer and simulation technology
o artificial Intelegent & interaction
ABM Theory o “ Agent-based modeling is a powerful simulation modeling technique
that has seen a number of applications in the last few years, including
applications to real-world business problems. In agent-based modeling
(ABM), a system is modeled as a collection of autonomous decision-
making entities called agents.” (Bonabeau 2002:7280)
o “Agent-based modeling is a bottom-up approach to understanding
systems which provides a powerful tool for analysing complex, non-
linear markets. The method involves creating artificial agents designed to
mimic the attributes and behaviours of their real-world counterparts. The
system’s macro-observable properties emerge as a consequence of these
attributes and behaviours and the interactions between them. The
simulation output may be potentially used for explanatory, exploratory
and predictive purposes”. (Twomey dan Cadman 2002:56)
o ”It is a methods for studying systems exhibiting the following two properties:
(1) the system is composed of interacting agents; and (2) the system
exhibits emergent properties, that is, properties arising from the interaction
of the agents that cannot be deduced simply by aggregating the properties
of the agents.” (Axelroad dan Tesfatsion 2005:3)
o ABM :
o Research / Experiment metodology
o bottom-up approach
o Interaction among individual behavior
o Computer simulation
o mimic the attributes and behaviours of their real-world
o interacting agents; and
o the system exhibits emergent properties
o yang dapat digunakan lagi sebagai alat bantu untuk
eksplanatori, eksploratori atau prediksi dalam mengambil
keputusan di dunia nyata.
The key feature of agent-based modeling Twomey & Cadman, 2002, Agent-based modelling of customer behaviour in the telecoms and media markets
o The term ``agent’’ in the context of business or economic modeling refers to real world objects such as people or firms.
o In the agent-based approach the focus turns to the properties of the individual agents.
o These agents are capable of displaying autonomous behavior such as reacting to external events as well as initiating activities. Of equal importance is the interaction of these agents with other agents.
o Involves a bottom-up approach to understanding a system’s behavior (e.g fish or bird group). o Traditional modeling usually takes a top-down approach in which certain
key aggregated variables are observed in the real world and then reconstructed in a model.
o Under this approach a modeler would observe the effects of say a price change on the number of consumers who purchased a product at an aggregated level. This would provide the basis for quantifying the strength of interaction in the model.
Agent Based Modeling Experiment Testfatsion, 2005, ACE Modeling Economies as Complex Adaptive Systems
o Modeler constructs a virtual world populated by various agent types (company, consumer, market, supplier, regulator)
o Modeler sets initial world conditions (consumer, market place)
o Modeler then steps back to observe how the world develops over time (no further intervention by the modeler is permitted)
o World events are driven by agent interactions
constructs a virtual world
sets initial world conditions
The world develops over time
Building an agent-based model Twomey & Cadman, 2002, Agent-based modelling of customer behaviour in the telecoms
and media markets
o Constructing a population of interacting elements or agents. These agents may be people, firms or other objects.
o Agents are designed to mimic their real life counterparts consists of: o Attributes (e.g. age, sex, preferences).
o Behaviours (e.g. actions based on decision making algorithms such a utility maximization).
o The attributes and behaviours may not only vary across agents in the model but they may also change within agents during the simulation. For example, an agent-based model may incorporate genetic algorithms or neural nets within the agents which allow the agents to learn throughout the simulation
o Once the agents are created and parameterised the model can then be run
o During and after the simulation run, output probes can provide information on any particular agent’s attributes and behaviours as well as the system’s macro-variables.
o One objective of the simulation is to provide the user with a good idea of how the real system would behave under the conditions portrayed in the model.
Perbandingan ABM dengan Model Kuantitatif
Secara Kuantitatif Pemodelan dengan Agent Based (ABM)
Model dibangun untuk
Model dibangun untuk mengungkapkan
permasalahan dengan pendekatan dari bawah ke
atas (bottom up approach), Twomey dan Cadman
Model dihasilkan dari
pengolahan data empirik
(seperti data hasil survey)
Model adalah langkah awal untuk menghasilkan
data empirik, simulasi yang dijalankan dengan
model akan menghasilkan data empirik ,
Axelroad dan Tesfatsion (2005:4)
Model yang dibuat untuk
Bukan model yang menyelesaikan masalah tetapi
agen-agen dalam model yang akan memecahkan
masalah yang dihadapi, Bonabeau (2002:7280)
Model yang dibuat adalah
hasil akhir dari penelitian
Model yang dibuat adalah langkah awal dari
penelitian, Bryson ett. all (2005:1)
Strengths of agent-based modeling o System assumptions, The emergent non-equilibrium, dynamical behaviour of a
system is usually one of the most interesting outputs of agentbased models.
o Realism. This allows us to undertake qualitative scenario exploration to investigate the structure or morphology of the system independently of the details.
o Natural representations. relatively easy to understand as they have a simple, structural correspondence between the ``target system’’ and the model representation. They are more intuitive and easier to understand than, say, a system of differential equations.
o Heterogeneity. ABMs also allow us to introduce a very high degree of heterogeneity (diversity) into our populations of agents. Traditional models ± to permit mathematical solutions
o Bounded rationality . Both limited information and limited abilities to process information may be explicitly incorporated into the model. Habit and social imitation may also be included.
o Communication and social networking. ability explicitly to incorporate communication among agents. Agents can, for example, ``talk’’, share information or imitate other agents in the population. This level of subtlety is usually outside the reach of traditional mathematical models, since social networks quickly make equation-based models so complex as to be insoluble.
o Object-orientated analysis, design and programming.
o Maintenance and refinement. It is reasonably easy to add new types of agents or new attributes or behaviours of agents without destroying earlier knowledge incorporated into the model
o + Ethical, parametric design.
Weaknesses of agent-based modelling
o Data problems. the potential lack of adequate data. This is not surprising since, as mentioned in the introduction, most quantitative research until now has concentrated on ``variable and correlation’’ models that do not cohere well with process-based simulation that is inherent in ABMs. This means that not only is it likely that new types of data are needed to be collected but even theories may need to be recast effectively to take account of the potentialities of agent-based simulation.
o Identifying rules of behaviors. Trying to capture the appropriate processes or mechanisms underlying the agents’ behavior may not be an easy task. However, as Hood (1998) points out, the flip side of this is that it forces us to be explicit about our assumptions and forces us to think about extracting the ``essence’’ of the problem.
o Programming skills. Any sophisticated, agent-based model requires programming in an object-orientated language such as Java. That is, it requires a level of computing skill beyond simple spreadsheet programming.
o Computational time. ABMs are computationally intensive, and although it is precisely because of the advances in computing power that we now have the possibility of desk-top agent-based modelling, there are still limits to the level of detail and number of agents that can be run in a simulation in a reasonable amount of time.
Designing an agent Hood, L. (1998), ``Agent based modelling’’, available at www.brs.gov.au/social_sciences/kyoto/hood2.html
o Low fidelity. o all the agents in the model have the same behaviour and intrinsic attributes.
o This situation would not even be categorised as an ABM by many practitioners.
o It is of interest for problems where the statistics of the collection of entities are of interest.
o This situation occurs in many physics and chemistry simulations (e.g. the molecular level simulation ofmaterial properties or drug design).
o because of their simpler agent details, usually much larger numbers of agents are employed in the simulations than in a typical ABM. For example, one of the largest astrophysics simulations ever performedconsisted of 150 million agents (stellar entities).
o Medium fidelity. o Here an observed distribution of the agents’ behaviour is used to ``calibrate’’ the model.
o This is a very useful middle ground to target for many applications where the tails of a distribution are of interest (e.g. the poorest 10 per cent, the richest 10 percent).
o An advantage of working at this level of detail is that it allows ups to capture some of the observed properties of the individual agents without having to resolve the internal workings of the agents (i.e. ``what makes them tick’’).
o High fidelity. o a proper attempt is made to capture the internal workings of the agents. This may include
trying to model, among other things, the beliefs, desires and intentions of the agent.
o At this level of fidelity we may also include an ability of the agent to adapt and learn, such that the agent’s behaviours and properties evolve over time as they learn about their environment and what actions lead to success or failure.
o At this level of fidelity we are thus capturing some notion of a mentalistic or cognitive agent.
o Swarn (berbasis bahasa C) o Bahan tentang Repast dapat didapat secara on line dari
o Repast (Recursive Porus Agent Simulation Toolkit:
berbasis Java) o Bahan tentang Repast dapat didapat secara on line dari
o Mason (Multi-Agent Simulator of Neighborhoods: untuk
kecepatan) o Bahan tentang Repast dapat didapat secara on line dari
o Netlogo (paling lengkap dokumentasi dan lebih praktis
digunakan) o Bahan tentang Repast dapat didapat secara on line dari
Proses Pembuatan ABM
Uji Validitas &
ABM Area of implementation (in Management)
Learning and the
There exists a broad range of algorithms which represent the learning process of
computational agents, e.g. genetic algorithms.
Norms are generated by interaction and in social settings. AXELROD (1997, 47) uses the
following definition “A norm exists in a given social setting to the extend that individuals
usually act in a certain way and are often punished when seen not to be acting in this way”.
of market processes
The major point in markets is the ability to perform self organisation. Some markets follow a
path dependency while others behave differently. Nearly every market can be investigated by
using agent-based simulations.
Economic networks play a crucial role in social and economic science. The formation of
transaction networks by strategically interacting agents takes the centre stage.
An organisation consists of a number of people which have an objective or performance
criterion that transcends the objectives of the individuals within the group (V. ZANDT, 1998).
In this sense organisations can be modelled by implementing agent-based models.
Automated markets This area is related to the Internet and to virtual markets. There is a number of profit oriented
research on the way with continuously growing implementations in products.
with real and
There are two main differences regarding experiments with real and computational agents:
The behaviour of computational agents is determined and known in advance while it is not
possible to know explicitly why real agents respectively human beings make a particular
choice. Performing both experiments in parallel could support the finding of insights.
Work with agent-based models needs computer and programming skills. There are
environments developed and still under construction which support application for non-skilled
researchers. These computational laboratories permit the study of systems of multiple
interacting agents by means of controlled and replicable experiments, e.g. Swarm or RePast.
CONTOH ABM dengan Netlogo
CONSUMER BEHAVIOUR MODEL
For VOICE MUSIC SMS (VMS)
VALUE ADDED SERVICE
AT GSM OPERATOR IN INDONESIA
o Secara konsep ABM diturunkan dari gabungan antar disiplin ilmu yang dikenal dengan konsep “Science complexity” istilah yang diangkap oleh Levin 1999[i].
o Secara alamiah konsep biologi dan ilmu sosial digabungkan sehingga menghasilkan gabungan yang kompleks yang dapat mengantisipasi sistem yang tidak linear, bisa mengatur diri sendiri, heterogen, bisa beradaptasi, ada feedback, dan dapat memunculkan perilaku.
o Ke semua gabungan ilmu tadi di implementasikan ke dalam suatu teknik computer dan software yang membuat kerangka kerja permodelan berbasis agen, yang merupakan hasil perkembangan teori komputer mulai dari artificial intelegent, neural network, dan pemrograman computer yang dapat berevolusi.
o [i] Lewin, R. (1999), Complexity: Life at the Edge of Chaos, University of Chicago Press, Chicago, IL.
ABM Cont. (Axelroad&Tesfatsion, 2005)
o Sistem yang terdiri dari agen-agen yang saling berinteraksi satu sama lainnya;
o Sistem yang memunculkan emergent properties yaitu sifat-sifat yang muncul dari interaksi antar agen yang tidak bisa di deduksi secara sederhana dengan mengumpulkan semua sifat dari agen-agen tersebut.
o Ketika interaksi antar agen merupakan kesatuan dengan pengalaman masa lalu, khususnya bila agen tadi secara terus-menerus beradaptasi terhadap pengalaman yang dihadapi,
o analis secara matematis biasanya sangat terbatas dan sangat sukar untuk memodelkan sistem
o [i] Axelrod, R dan Tesfatsion, L., 2005, “A guide for Newcomers to Agent-based Modeling in the Social Sciences”, University of Michigan