Inductive Learning - Advanced Artificial Intelligence - page 295

of λ 1 λ 2 λ 3 are (where T in , E S , E are tuple interval, error-classification ratio and

expectation of generating time respectively):

η

E × E

S η

λ = 0.1 × T , λ = 2, λ =

1 in 2 3

2

Definition 7.25 An algorithm is a optimum algorithm Γ if and only if it meets
optimal index ζ and makes λ 3 minimum.

Definition 7.26 The background handling report REP is a text file ErrRep.txt.

when BSDT cannot find optimum algorithm Γ in appointed path, it will record
error information, and inform manager to handle it.

Definition 7.27 In BSDT algorithm the register procedure should do following

tasks:

a

(1) Store name of an algorithm N into algorithm index table AI, and store

algorithm itself in directory appointed by BSDT.
(2) Store name of an example set N

es

into example index table EI, and store
classical case table TET in classical case base.

7.8.5 Bias shift based decision tree learning algorithm BSDT

BSDT algorithm firstly call seek optimum algorithm SM to get optimum
algorithm Γ . Then seek the algorithm in system directory and current working
directory. If no algorithm is found, it should be eliminated from case base and
algorithm index table, in case fail to find next time.
In fact, seek optimum algorithm SM is a procedure using case-based
reasoning. As for a given primitive training set, we firstly extract its retrieval
information tuple θ, then search optimum algorithm meet optimal index ζ in
case base. When case base is empty or SM algorithm fails, take GSD as default
selection of optimum algorithm.
Since inference in algorithm running should be decreased, BSDT journal can
be taken as a means for manager to intervene system.

Algorithm 7.8 Bias Shift Based Decision Tree Learning Algorithm BSDT 。

Input: primitive training example set;
Output: target knowledge base;