The event extraction mechanism takes these PEPs and applies them to the training instances. The event extraction algorithm is shown in figure 5.4.
Figure 5.4: Event extraction algorithm
The algorithm takes each stream, and each channel in each stream, and
then applies each appropriate PEP to that channel. The results are
then added to the event list E for later use. 
is the
finding function of the PEP 
.
Not all PEPs are appropriate for all channels, so we must use our
domain knowledge to decide which PEPs should be applied to which
channels. Thus 
returns true if PEP p can be
appropriately applied to channel c. For example, if we know that a
channel is highly noisy, it doesn't make sense to apply a local
maximum PEP, since it's not likely to pick up salient features of the
data so much as random noise. Similarly, it does not make sense to
apply a ``delta'' PEP to a continuous channel, or a ``straight line''
approximation to a discrete channel.
The result of the operation is that E now holds a set of tuples, each tuple consisting of some identification as to which stream and which channel it belongs to.
If we were to apply this to the Blues and Reds task, we would get the data shown in table 5.2.
Table 5.2: Event extraction applied to the Blues and Reds domain