Information Technology Reference
In-Depth Information
Algorithm 1. PMEA (VP)
1 VP RBP;
// Rule-based expert system
2 for i ∈ N do
3
PlayerModel PlayGameOn(VP);
// Player modeling (PM)
VP EA(PlayerModel,VP);
// Evolutionary optimization (EA)
4
end for
5
return VP;
6
Fig. 2. Example of an arbitrary encoding (left), and extended answer matrix (right)
situation in the environment. An individual is represented as a vector v of k cells
where k is the number of different situations in which the agent can be in the
game, and v [ i ](for0
i<k ) contains the action to be taken in situation i .In
other words, assuming m perceptions, where the perception p j (for 0
j
m
1)
can have k j
possible values, then k = k 0
k 1
...
k m− 1 and the cell:
v [ e m− 1 + e m− 2
k m− 1 + e m− 3
( k m− 1
k m− 2 )+ e m− 4
( k m− 1
k m− 2
k m− 3 )+ ...... ]
contains the action to be executed by a specific unit when its perceptions
p 0 ,p 1 ,...,p m− 1 have the values e 0 ,e 1 ,...,e m− 1 respectively. As indicated in
Section 2, the state of a unit in our wRTS game is determined by three boolean
perceptions and its energy level (with 3 different values) so that the encoding of
an individual consists of a vector with 24 genes; this vector will be called answer
matrix , and each cell in the vector will contain an action. Remember that 6 pos-
sible actions can be executed in our wRTS and thus the search space (i.e., the
number of different strategies that can be generated from this representation)
is 6 24 . Figure 2(left) displays an example of a possible encoding. The optimal
solution (if it exists) would be that strategy which always select the best action
to be executed for the agents under all possible environmental conditions. In
fact, this vast search space makes this problem impracticable for exact methods
and justifies the use of evolutionary algorithms.
3.3
On-Line Phase: User Modeling
The player behavior will be modeled as a ruled-based strategy encoded as ex-
plained above. This process requires collecting, during the execution of the game
Search WWH ::




Custom Search