Information Technology Reference
In-Depth Information
9. If above conditions are not met, then go to step 1;
10. Call algorithm MEA to train example subset and maintain attribute set;
11. Call algorithm STCH to store concept level tree of
A
i
into concept level database D;
12. Go to step 1.
7.8.4
Procedure bias shift
Various decision tree learning algorithms have their own advantage, and the
basic point of our proposed decision tree learning algorithm based on bias shift is
employ their advantages. At the same time, learning algorithm has complicated
relation with their learning task and attributes of training set, such as size,
dimension, domain, etc. we cannot use simple controlling branch sentence to
realize the selection of seek optimum algorithm. Therefore, we introduce the
concept of two-level decision tree learning algorithm based on bias transform
(see Figure 7.8).
The algorithm is designed based on the idea of two-level and multiple
strategies. Two-level learning places the focus on: the first level is case-based
reasoning used to select the most proper algorithm to solve primitive training
example set from various decision tree learning algorithms with different
evaluation standard, adaptive domain, size of training examples; the second
learning task is used to construct classifier, that is, using selected decision tree
algorithms, infer classification rules of decision tree representative form from
cases without order and rule.
Case-based reasoning is a strategy that gets source case in memory from
prompt of target case, and guides target case solving from source cases. Here,
target cases are generated from classical case base of various decision tree
algorithms. As for a given primitive training instance subset, we firstly extract its
retrieval information tuple θ to retrieval in case base. Similarity in retrieval
procedure meets optimal index
ζ
in definition 7.24.
Multi-strategy learning is not constrained in one learning algorithm. It also
provides mechanism to introduce new algorithms and classical examples. The
mechanism provides a seamless link among original register algorithm, new
classical case set and original register classical case set. The mechanism is
realized by interface of human-machine interaction and classical case base
maintain algorithm.
b
1
ϖ
]
T
Definition 7.14
represents relativity
between various attributes of primitive training set and learning task, where
ϖ
is
number of attribute of primitive training example set.
Bias coefficient C
= [ C
,
…
, C
