
50 Chapter 4
If use of every contiguous group of points results in what is
considered to be an excessive number of training samples, it is
reasonable to use a subset. However, extreme care should be taken if
a deterministic method is used to choose the subset. Straight
decimation (selection of every nth group) can easily cause bias in the
training set. It is not uncommon for patterns to vary in cycles.
Deterministic sampling can emphasize some patterns, while missing
others. A far better approach is to select training samples randomly
from the training series.
Figure 4.1 shows the predictors being contiguous points (lags
from 0 - 5). Many