Out: October 2, 2009
Due: October 13 with competitions Monday/Tuesday/Thursday
Subproject 3: Improving Your Current Agent and Planning for Future Agents
The subproject consists of two parts. In the first part you improve the current agent by paying attention to the problem providing task (ProvideTask.java). For example, when an agent accepts a TBall challenge using relation 22, it will receive a problem using only relation 22 and the offerer of the TBall challenge will try to find a problem where only a small fraction of the constraints can be satisfied. How small can this fraction be? We know that it it cannot be smaller than 4/9 because we already determined that there must be an assignment satisfying the fraction 4/9. But it turns out that we can create problems where the best assignment is arbitrarily close to 4/9. We discussed this construction in class. It was about making your challenges as attractive as possible within the constraints defined by the configuration file. We analyzed this requirement using binomial coefficients.
Next we focus on future agents that we want to implement in the future. In class on September 25, I outlined the transition from a money-based artificial market to a new reputation/confidence-based artificial market that has the flavor of an artificial scientific community. We could call it an artificial science market. I have summarized the artificial science market in the first section of: Vision for New Requirements The other sections are the same old requirements document you have seen before.
It is important for software developers to engage with their users and help them to formulate the detailed requirements based on the vision of the users. In part 2 you write a requirements document that proposes a specific mechanism to implement the axioms described in the above vision document. This mechanism should be similar to the current game. We want to change the requirements minimally.
In part 3, we consider a new kind of belief which we call secret beliefs.
In your team of two students
play a few rounds of
SDG/secret as described in
Use MAXCSP as optimization problem and play a TBall game (only
one relation defining the predicate of a belief).
Note how the new beliefs call for different optimization algorithms
and a different way to construct problems that hide secrets.
Turn in the history of the game you played.
Part 1: Points: Based on performance in competitions
Create an improved agent that creates appropriately hard problems for other agents to solve. Make your challenges more attractive to other agents, but don't make them so attractive that you loose money/ reputation.
Part 2: 15 points
Create a 2 page document that describes the impact that the transition to the reputation-confidence based market will have on your current TBall player. Suggest new rules thnat the administrator shoult enforce.
Part 3: 8 points
Turn in the history of the secret game you played. Design a class dictionary to define the history, generate a parser and parse your history without error before you turn it in.