Game Development Reference
In-Depth Information
example, the face value of $20 didn't mean the same to everyone. The 20 is a concrete
value; the utility that people put on those $20 is an abstract rating. We will revisit
this later.
C OMBINING M ULTIPLE C RITERIA
The important task above was to decide on the range (and, by association, the gran-
ularity) we were going to use to score our criterion. When all we are concerned
about is a single criterion, we have plenty of flexibility in how we do this. However,
when we plan on using one utility score in conjunction with others, it behooves us
to keep the whole picture in mind.
N ORMALIZING
One way of making the job of comparing, contrasting, and combining disparate
numbers easier is to use the same scale for all of them. If, for example, we are track-
ing satisfaction ratings for various things, we want to place all satisfaction ratings on
the same scale. In Chapter 9, we examined the hedonic calculus factors that could
be involved in a decision on where to go out for dinner. While some of the factors
were concrete (such as cost, travel time, and wait time), some of them were subjec-
tive satisfaction ratings. By making sure that we were rating all satisfaction-based
values on the same scale, we can ensure that we have little difficulty later on.
Technically, the term normalization refers to the process of stretching or
shrinking a range by a normalizing constant so that it fills the space from 0 to 1.
This is a very common practice in probability math. While we may not be using it for
exactly the same purpose, the rationale behind normalizing subjective factors is
similar: put everything into one, homogenous template. For our purposes, this
template doesn't have to be 0 to 1. As we have suggested above, we could use the in-
tegers 0 to 100 (pseudo-percentages), 0 to 255 (fill a byte variable), or something
as simple as 0 to 10 such as what the pain scale uses. There is one very significant ad-
vantage to using 0 to 1, however: when we multiply factors that range from 0 to 1
together, the product is still in the range of 0 to 1. That is remarkably handy when
producing weighted sums or weighted averages, for instance. Regardless of the range,
however, the main consideration is the granularity that we require for the value.
As we mentioned above, we can normalize concrete values as well. We can do
this linearly or through a function such as the ones we covered in Chapter 10.
 
 
Search WWH ::




Custom Search