Game Development Reference
In-Depth Information
While that approach worked for our sample data, imagine that our scores had
turned out more like what we see with the gray bars in Figure 16.1. In that example,
there is a large jump between the score of the first two and the rest of the options.
We can no longer consider these eight options as being equal. That hamstrings our
idea of selecting randomly from that list of eight.
FIGURE 16.1
If there is a significant difference between the scores of our
top
n
selections, we should not treat them equally.
One possible solution to this problem is to reduce our cutoff to two possibilities
instead of eight. However, that almost negates the purpose of expanding it in the
first place. Also, we have no way to predict where that cusp will be on any iteration
of this process. Looking at our original data (black bars), we could have included
the first
12
records before we saw a significant jump in the scores.
To continue to provide some variety in our choices, we would like to continue
drawing from the top eight selections (or, more generally, the top 25% of the
possible selections). What we do
not
want to do, however, is to treat all eight pos-
sibilities as equal.
Thankfully, our scoring system provided us with something other than an or-
dinal ranking of the possibilities. The very fact that the scores vary from one option
to the next reminds us that these differences are proportional. In layman's speak,
some options are “more better� than the options that come after them. We need a
way to leverage these proportional differences.
