Wonders of Seabed: Difficulty Evaluation of Management Games Using Neural Network - Technologies and Applications of Artificial Intelligence - page 172

where t is the current step, ʷ is the learning rate (0.3 in our system), and ʔ w
is the change of the weight vector.
After finishing all the three steps, our system learns the settings about the
tax, maintenance fee, and so on. And then the process is repeated until the
termination condition is satisfied. During the iterations, we record the total
amount of money that would be earned. This total amount of money indicates
whether or not the game rules are too dicult or not for the player. Our system
can then adjust thresholds when necessary to suit for the player's expectation
achievement level.
5 Result and Discussion
We performed experiments to evaluate the game di culty levels in several ex-
amples. The number of the evaluation iterations of the neural network was set
to 100. Different game time limits were employed. A game time limit is the time
duration for playing a game. We recorded the average total incomes, as shown in
Table 1. One month game time is set as 30 seconds in real life. Our observations
are stated as follows:
1. In Table 1(A), for the first 20 months, the total income is always negative.
From 15 months to 20 months, the total income is increasing because the
neural network learned better than before. So we could say that players start
to know how to balance the parameters to get more income starting from the
20-th month. If we set the stage goal to be 20,000 dollar, a player needs 40
months to clear the stage. In other words, in real life time, the player takes
20 minutes to clear the stage. Based on our experimental results, we can use
the neural network to control how long a player needs to clear a stage.
Tabl e 1. Experiment Results
A. ( T tax , T maintain , T cele , T human ) = (750,250,100,50)
Time to play (month) 10 15 20 30 40
Average total income -4804 -5125 -3241 5392 20670
B. ( T tax , T maintain , T cele , T human ) = (750,250,500,500)
Time to play (month) 10 15 20 30 40
Average total income -5927 -6340 -3515 1592 16213
C. ( T tax , T maintain , T cele , T human ) = (850,100,500,500)
Time to play (month) 10 15 20 30 40
Average total income -694 1224 6359 20350 40583
D. ( T tax , T maintain , T cele , T human ) = (850,100,100,50)
Time to play (month) 10 15 20 30 40
Average total income -159 3301 8507 22027 44412
E. ( T tax , T maintain , T cele , T human ) = (600,400,400,400)
i e to lay ( o t ) 10 15 20 30 40 50 60 70 80 90 100
Average total income -10428 -13219 -16452 -18420 -16153 -13271 -7164 1654 5427 7278 21462