The Model 

The El Farol model was described by Brian Arthur. This implementation of the model based on the NetLogo version, which may give a good opportunity to compare the languages.

The model is about a bar close to the Santa Fé Institute where every Thursday workers decide to visit the pub or stay at home. Unfortunately, the bar is a bit small and the time spent there only worth it if less than the 60% of the population goes to the bar due to overcrowdedness. However, if the bar was overcrowded, the people stayed at home had a good time.

Agents have to decide their actions at the same time, and they do not know anything about the opinions of the other agents. The only information available for them is the last weeks' occupancies. In this way they cannot see if it is worth visiting the bar before their decision.

An interesting fact of the problem is that there is no deterministic strategy for the agents. If each agent use the same strategy and decide in the same way, it would mean they do the same actions as the others and come to the same (bad) decision. For instance, if a strategy implies that the bar will not be crowded, everyone (following the same strategy) goes to the bar and it becomes crowded. In the other case, if strategy implies that the bar will be crowded, nobody will visit the bar andit remains empty.

The model is closely related to minority games.

The article describing this implementation of the model can be downloaded here.
The exported Fables project can be downloaded here.
To import the model, select File --> Import and chose the Fables / Fables Project wizard. 

How does it work?

There is a fixed number of agents deciding each simulation step if they whether go to the bar or not. They have several random strategies and try to predict the attendance level of the bar separately.

• In each turn if the bar was overcrowded, agents stayed home made a good decision.
• If the bar was not overcrowded, agents went to the bar made a good decision.

The default tolerance level of the bar crowdedness is 60% by default. 

Modifying the model

There are several extensions of the model, including genetic algorithm and reinforced learning. Try to implement other rules for the prediction.

References

Arthur, W.B.: Inductive Reasoning, Bounded Rationality and the Bar Problem. Working Papers 94-03-014. Santa Fe Institute, New Mexico, USA, 1994.

El Farol Bar problem From Wikipedia, the free encyclopedia

Rand, W. and Wilensky, U. (2007). NetLogo El Farol model. Center for Connected Learning and Computer-Based Modeling, Northwestern University, Evanston, IL.

Legéndi, R., Gulyás, L., Bocsi, R. and Máhr, T.: Modeling Autonomous Adaptive Agents with Functional Language for Simulations. Lecture Notes in Computer Science, 2009, Volume 5816, Progress in Artificial Intelligence, Pages 449-460