I am a computer scientist trying to build agent based operations research and social simulation models using UML statecharts. I start my conceptualization by thinking about who are the actors to be considered and what kind of states they might be in and what would trigger changes in states. More details can be found in the attached excerpt:
I am still trying to understand better how other people (from ecology or economics or social simulation) start their modelling process (and why do they start that way).
I heard from my colleagues in ecology that they would start with a Brownian motion model and then add some attraction/repulsion models and perhaps some other theoretical models to fine tune the Brownian motion model. We were discussing a model of a chicken farm where they were studying where best to locate the food sources.
Can you tell me how do you usually start your modelling activities? I would also be interested to hear if you consider your model to be object oriented (either at modelling level or at implementation level or at both levels).
I wish I could give you more detailed advise. I am currently following a free online course on complexity (http://www.complexityexplorer.org/) that you can try and see if it helps you move forward wit agent-based modelling. Maybe you can also explore the Santa Fe Institute site (http://www.santafe.edu/) and contact some faculty member for guidance...
I don't think it is a good idea to look at other research domains as each domain brings its own specificities that often don't have equivalence in others. In my lectures on modelling, I always try to convince students that, in modelling, the decisive point (and the first one to consider) is the precise question you want to answer with your model. This is the main element that can help you deciding what you have to put in your model and what you can remove from it. Once you have the question, you can choose the kind of model that is best suited to answer it and, then, precisely tune your model.
Considering the question on OO, I program my ABM models in C++ but with an extensive use of pointers... Remember that agent-based models can be highly demanding in terms of computing resources (CPU/Memory). Thus you'd better choose a very efficient language and programming structures. The loss of time due to a more time-consuming development is negligible when compared to the gain during the experimental phase.
To conclude, I would like to recommend the last book from Railsback and Grimm that is very informative and also a pleasure to read (http://www.railsback-grimm-abm-book.com/)...
Here you can find a selection of tutorials for NetLogo agent-based modeling: http://www.openabm.org/page/modeling-tutorials. Maybe you can identify a heuristic approach by exploring the Models Library.
A lot of good ideas here, and I can second the value of Railsback and Grimm's textbook, as well as the use of NetLogo as a potential platform for generating an IBM.
Another good book is found here (by the same authors) that is more about the theory of IBMs rather than the direct application within NetLogo - http://press.princeton.edu/titles/8108.html. I consider it a must-read and has really helped me think about how to go about building an IBM (and that it really has to be specific to your research question to prevent over-complexity).
I would also suggest looking at the ODD Protocol (citations below). While it is meant as a means for describing IBMs within the literature (to allow for understanding and reproducible results across IBMs) I also find it helpful in planning the building of an IBM as well.
Grimm, V., Berger, U., Bastiansen, F., Eliassen, S., Ginot, V., Giske, J., … DeAngelis, D. L. (2006). A standard protocol for describing individual-based and agent-based models. Ecological Modelling, 198(1-2), 115–126. doi:10.1016/j.ecolmodel.2006.04.023
Grimm, V., Berger, U., DeAngelis, D. L., Polhill, J. G., Giske, J., & Railsback, S. F. (2010). The ODD protocol: A review and first update. Ecological Modelling, 221(23), 2760–2768. doi:10.1016/j.ecolmodel.2010.08.019
In short, the model design is going to be very specific to your research questions, and therefore the amount of complexity needed. Hopefully these resources help.
Hi !
This is a very large and fundamental question...with no accurate answer. In fact there is no precise and efficient protocol (theorem) to follow which can guarantee a successful achievement of the job. But we can draw roughly a scheme which can help in designing the model and avoid some errors and/or dead-ends.
1. Guillaume is right (hello Guillaume !). The fundamental question is "what is your goal" ? What answer to what question do you intend to obtain with your model ?
2. Collect all data and available knowledge about the system (the real system).
3. Considering the points above, decide which processes must be kept and which processes can be ignored (removed) into your model. This step is always delicate and can strongly influence the pertinence of your results.
4. For each of the retained processes (movements, death, birth, prey capture, foraging rules ....) decide the level of detail at which they must be implemented. As a result of this decide the granularity of time (second, minute, hours, day, season ???)
5. Draw the conceptual model. It is something which looks like a flow-chart including time management, indicating the place of the processes at which they will act within a temporal cycle of the model.
6. If you retain an OO-language (C++, Java...), draw an ULM chart. This will help in implementing classes and their relations, methods...
7. Implement your model (write the code), starting with the generic classes.
8. Try to run (and debug) your model with a set of initial conditions. Does the result are satisfactory from a qualitative point of view ? If no, return to step 3, otherwise enter in the validation process and sensitivity analysis...
Lastly, I have a little secret to tell you : to my knowledge, nobody follows exactly this kind of receipt. Because modeling is not a science, it is something like an art (in the sense or "artisan"). The experience and skill of the modeler are fundamental keys in building an ABM of high quality.
Good Luck !
I also agree with Nathan Furey's argument to have a look on the ODD protocol to describe AMBs - also as a means to structure the development work of an ABM. I also used it for a structured analysis of ABMs, and found the ODD protocol very helpful for this task. Find the work here: https://www.researchgate.net/publication/262891441_Analysis_of_Agent-based_Tourism_Demand_Modeling?ev=prf_pub
Thesis Analysis of Agent-based Tourism Demand Modeling
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India