LBR(for Lazy Bayesian Rules) is a Bayesian classifier that defers all process-
ing to classification time. For each test instance it selects a set of attributes for
which the independence assumption should not be made; the others are treated
as independent of each other given the class and the selected set of attributes.
It works well for small test sets (Zheng and Webb 2000).
LW Lis a general algorithm for locally weighted learning. It assigns weights
using an instance-based method and builds a classifier from the weighted
instances. The classifier is selected in LW L’s object editor: a good choice is Naïve
Bayes for classification problems and linear regression for regression problems
(Section 6.5, pages 251–253). You can set the number of neighbors used, which
determines the kernel bandwidth, and the kernel shape to use for weighting—
linear, inverse, or Gaussian. Attribute normalization is turned on by default.Miscellaneous classifiers
The misc. category includes two simple classifiers that were mentioned at the
end of Section 4.7 (page 136).Hyperpipes,for discrete classification problems,
records the range of values observed in the training data for each attribute and
category and works out which ranges contain the attribute values of a test in-
stance, choosing the category with the largest number of correct ranges.VFI
(voting feature intervals)constructs intervals around each class by discretizing
numeric attributes and using point intervals for nominal ones, records class
counts for each interval on each attribute, and classifies test instances by voting
(Demiroz and Guvenir 1997). A simple attribute weighting scheme assigns
higher weight to more confident intervals, where confidence is a function of
entropy.VFIis faster than Naïve Bayes but slower than hyperpipes.Neither
method can handle missing values.10.5 Metalearning algorithms
Metalearning algorithms, listed in Table 10.6, take classifiers and turn them into
more powerful learners. One parameter specifies the base classifier; others
specify the number of iterations for schemes such as bagging and boosting and
an initial seed for the random number generator. We already met FilteredClas-
sifierin Section 10.3: it runs a classifier on data that has been passed through a
filter, which is a parameter. The filter’s own parameters are based exclusively on
the training data, which is the appropriate way to apply a supervised filter to
test data.Bagging and randomization
Baggingbags a classifier to reduce variance (Section 7.5, page 316). This imple-
mentation works for both classification and regression, depending on the base414 CHAPTER 10 | THE EXPLORER