Information Technology Reference
In-Depth Information
kinds of different profit mechanism: profit-punishment,
p
¬
0; just considering
profit,
'
¬
0. The first situation reflects
the environment, showing that the result of the system is incorrect. The second
situation responds to the environment, and there is no difference before correct
action is generated. The third situation then reflects a general case, where there is
an action getting the maximal profit and the others get the same profit. In step 2,
e is regarded as the cost of each classifier in [
p
and
R
' are all equal to 0; profit-profit,
R
A
] in order to obtain the profit. The
total profit of [
] is assigned by distribution function D, and the simplest method
is to share R with each classifier.
The discovery component is based on Holland's genetic algorithm, using
reproduction, crossover and elimination. Boole's genetic algorithm is presented
as follows.
Algorithm 13.6
Boole's genetic algorithm
A
1. Select a classifier
] according to probability and intensity.
2. Similar to step 1, select classifier
C
1
from [
P
C
2
according to probability
χ
, and then
cross
C
2
with
C
1
; then select one result as offspring, and the other one is
discarded.
3. If step 2 is not finished, reproduce
C
1
so as to generate offspring.
4. Apply mutation operation to generated offspring, that is, change allele of each
individual with probability
Ⱥ
.
5. If each parent's intensity is reduced by one third through mutation operation,
then let offspring's initial intensity is equal to the sum of the reduced intensity
of parents, else that of
1
is reduced by one half and initial intensity of
offspring is equal to the reduced scale.
6. Add offspring to [
C
].
7. Delete the classifier with the minimal intensity from [
P
P
].
In each executing circle, the algorithm is called with probability
Ⱦ
. This
algorithm is basically the same as the normative genetic algorithm. The
difference between this algorithm and the normative genetic algorithm lies in the
fact that the former can generate only one offspring when it is called at a time,
while the latter usually generates many offspring.
In addition, discovery component also uses another mechanism---creation.
Namely, if [
] is empty in executing circle, then a new classifier is created. The
name of the created classifier is a replication of currently input string. The
probability of inserting # is equal to the proportion of # in [
M
P
]. In order to get
