Results for Exercise Sheet 11 (Margin size of Naive Bayes)
Please add your row to the table below, following the examples already there. The number of outliers is the number of points on the "wrong" side of the hyperplane implicitly computed by Naive Bayes (see Exercise 11.1). The margin size is the minium distance of the points from one class to the hyperplan + the minimum distance of the points from the other class to the hyperplane, ignoring all outliers.
NOTE: the first line below the headers provides the results for an SVM classifier, as explained and demonstrated in the lecture. Note that the SVM classifier manages to completely linearly separate the training set in both cases, and still the margin size is larger. However, the precision increase only for one of the data sets, so it's not clear how much a larger margin size helps prediction accuracy.
Name |
Actor vs. Politician |
Singer vs. Songwriter |
||||
#outliers |
margin size |
precision |
#outliers |
margin size |
precision |
|
SVM |
0 |
0.66 |
98% |
0 |
0.45 |
92% |
Björn |
5 |
0.01 |
98% |
8 |
0.01 |
89% |
MartinM |
170 |
0.02 |
99% |
818 |
0.001 |
96% |
José |
310 |
0.001 |
98% |
1267 |
0.0004 |
88% |
Freya & Wera |
6 |
0.02 |
98% |
10 |
0.001 |
57% |