41
There are definitely some cases where you don’t want a feature being included in the
calculation of a target feature. The most important case is when the target feature is a feature
itself. Indeed, all of the target features are also features, and in some cases, this is appropriate.
It is not appropriate however, to have a target feature use itself in a calculation. The reason this
is a problem is twofold. Let us consider the simple cut as an example. If the feature is cut and
the target feature is cut then you have selections of cut and no cut as well as options of cut and
no cut, respectively. What will happen though, is that every time feature is cut, target feature
will also be cut. And similarly for no cut. This could end up skewing values greatly. The
probability of a cut when there is no cut is effectively 0 percent. You don’t want to multiply zero
times other probabilities because you will most certainly end up with zero. For this reason, it is
important to skip a feature when it matches the target feature.
There are certainly cases where you would want to use a target feature as a feature though,
and that’s why they are included. For example, when deciding on what the shot type might be,
whether it was a cut or not on that exact same line is actually pretty important. As a matter of
fact, if there is no cut, the likelihood that the shot type would be the same on this line as the
last line is pretty high, since you would expect that unless the camera moved or a character
moved, things that do happen but not necessarily frequently, then the shot would be the same.
On the other hand, if there is a cut, it might still be the same shot type on this line, like if you
were cutting back and forth between two close-ups, but it is also just as likely that it would be a
different shot type. Likewise, once you know the shot type, you might want to use that for the
clean type. These are reasons why target features are included as features. Furthermore, each