Game Development Reference
In-Depth Information
Imagine for a moment a character in a role-playing game (RPG). The character
may have 4 different weapons to select from just as our agent above did. In addition
to that, however, our character may have 3 offensive magic items, 2 defensive magic
items, 8 different types of potions, 20 spells (some offensive, some defensive), and
a number of non-combat options. If he was also facing multiple opponents whom
he could attack (in any of those methods) and had multiple allies whom he could
assist, the number of possible actions would increase geometrically.
Using the methods we have discussed, we can score each of the possibilities in
the same manner, weight them appropriately, and easily select a weighted random
result from among them—
even if there are hundreds of possible actions
. We do not
need to be horrified by the prospect of our agents having that flexibility. We can
give them the freedom to choose and feel comfortable that they will exhibit that
precious balance of rationality and variety.
We made one very significant assumption in the examples in this chapter: Scores
can be easily converted to weights. In both methods of converting the Dude assault
score to a weight, we proceeded on the premise that the distribution of scores
would provide proper weights for us with a minimum of manipulation. In this
case, the outcome seemed reasonable. However, that may not always be the case.
(For that matter, we don't even know that this
is
the case in what we did above.)
If we think back to the gymnastics and skating examples, the scoring systems
that they use have all of the contestants compressed into a very small range. There
is little difference between them. Furthermore, the 0.05 difference between 9.85 and
9.90 may not mean the same as the 0.05 difference between 9.45 and 9.50. We just
don't know what those scores
mean
. Because of the lack of differentiation between
the scores, it might be difficult to convert this scoring system into a weighted selec-
tion system that represents the range from excellent to horrible.
We can say the same for our abstract score for the dudes. As we processed that
score, we went from very concrete numbers like “damage per second� and “dis-
tance to detonator� to more abstract “threat ratios.� Unlike the concrete values, our
final score is not something that we can look at and mentally decipher. Thankfully,
it seems that this system worked properly. Testing and observation may lead us to
adjust the formula that we use to convert the scores to weights, but for the most
part we are in the right neighborhood.
