Page 100 - Data Science Algorithms in a Week

84 Fred K. Gruber

Density: Level 2 Cold vs Warm: Level 2 Density: Level 3 Cold vs Warm: Level 1
Pointalized: Level 2 Saturation: Level 3 Pointalized: Level 2 Saturation: Level 1
Light/Dark: Level 3 Motion blur: Level 2 Light/Dark: Level 2 Motion blur: Level 1
BKG: Level 1 BKG: Level 2
Figure 9. Images with features 2223321 and 3121212, respectively.

As an illustration, typical images for different values of these features are shown in
Figure 7 to Figure 9.
The response of each individual is an independent dataset. Rabelo (2001) compares
the performance of several learning algorithms on this collection of images.

Implementation Details

The support vector machine is based on a modified version of LIBSVM (Chang and
Lin, 2001) while the genetic algorithm implementation was written from the ground up in
C++ and compiled in Visual C++ .NET. In the following, we describe more details about
the genetic algorithm implementation.

Representation

Each individual is represented as a binary string that encodes five variables (see
Figure 10):

 The first 16 bits represents the cost or penalty value, C. It is scaled from 0.01 to
1000.
 The next 16 bits represents the width of the Gaussian kernel,  , scaled from
0.0001 to 1000.
 The next 2 bits represents 4 possible values for the degree d : from 2 to 5
 The next 16 bits represents the  parameter, which controls the percentage of

polynomial and Gaussian kernel. It was scaled from 0 to 1.
r
 Finally, the last parameter is the value, which determines whether we use a
complete polynomial or not.