Information Technology Reference
In-Depth Information
Fig. 2. Example of a random probing with the comutation of μ RP.Max
values) are created based upon these values. The first set is denoted M X and contains
all fitness values of the randomly selected points. The quartiles for the distribution of
the fitness values are computed and the values between the upper or lower quartile
are joined into the second set. This set is denoted M iqr . The third set M LU consists
of the fitness values which are between a lower and upper boundary L and U .These
boundaries are defined by L = Q 1 + 2 ( Q M −Q 1 ) and U = Q M + 2 ( Q 3 −Q M ) where
Q M denotes the median and Q 1 ,Q 3 denote the lower and upper quartile of M X .For
each set M X
M LU the number of elements is determined.
The feature Random Probing Min μ RP.Min is calculated based on the linear
model that fits the relationship between the number of values and the minimum fitness
values in each set. The straight line of the model is divided by the interquartile range
of M X . Similar to this the feature Random Probing Max is based on the slope
of the straight line that describes the relationship of the number of elements and the
maximum value of each set M X
M iqr
M LU (see figure 2). The slope divided
by the interquartile range of M X denoted by μ RP.Max is the second feature of this
group. Finally, for the feature Random Probing Range denoted by μ RP.Range the
spread, that is the difference between the maximum and the minimum value, in each set
is computed. As for the other features the slope is divided by the interquartile range of
M X . All features of this group are computed based on the fitness values of the randomly
selected points. For each point the objective function is evaluated once, hence, k = 100
evaluations are necessary for Random Probing .
M iqr
3.2
Incremental Probing
In contrast to the features of the previous group, Incremental Probing is computed by
the fitness values of the particle positions which are located in a defined distance to
a pivotal element which we choose from the feature group above. In order to calcu-
late the relevant fitness values, the position of a randomly selected pivot element is
 
Search WWH ::




Custom Search