anytime algorithms for learning anytime classifiers saher ... - Technion
anytime algorithms for learning anytime classifiers saher ... - Technion
anytime algorithms for learning anytime classifiers saher ... - Technion
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<strong>Technion</strong> - Computer Science Department - Ph.D. Thesis PHD-2008-12 - 2008<br />
Average Size<br />
Average Accuracy<br />
180<br />
160<br />
140<br />
120<br />
100<br />
80<br />
60<br />
40<br />
20<br />
IIDT(0.1)<br />
Skewing<br />
Sequential Skewing<br />
0<br />
0 1 2 3 4 5 6 7<br />
100<br />
90<br />
80<br />
70<br />
60<br />
50<br />
Time [sec]<br />
IIDT(0.1)<br />
Skewing<br />
Sequential Skewing<br />
Bagging-ID3<br />
Bagging-LSID3<br />
0 1 2 3 4 5 6 7<br />
Time [sec]<br />
Figure 3.34: Anytime behavior of modern learners on the Multiplexer-20 dataset<br />
its results separately later in this section.<br />
The graphs <strong>for</strong> the first 2 problems, which are known to be hard, show that<br />
IIDT clearly outper<strong>for</strong>ms the other methods both in terms of tree size and accuracy.<br />
In both cases IIDT reaches almost perfect accuracy (99%), while bagging-<br />
ID3 and skewing topped at 55% <strong>for</strong> the first problem and 75% <strong>for</strong> the second.<br />
The inferior per<strong>for</strong>mance of bagging-ID3 on the XOR-5 and Multiplexer-20<br />
tasks is not surprising. The trees that <strong>for</strong>m the committee were induced greedily<br />
and hence could not discover these difficult concepts, even when they were combined.<br />
Similar results were obtained when running bagging over C4.5 and RTG.<br />
However, when our LSID3(r = 1) was used as a base learner, per<strong>for</strong>mance was<br />
significantly better than that of the greedy committees. Still, IIDT per<strong>for</strong>med<br />
significantly better than bagging-LSID3, indicating that <strong>for</strong> difficult concepts, it<br />
is better to invest more resources <strong>for</strong> improving a single tree than <strong>for</strong> adding more<br />
trees of lower quality to the committee.<br />
67