Database Reference
In-Depth Information
Space Shuttle
Sine Cubed
Noisy Sine Cubed
1
1
1
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
0
0
0
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
ECG
Manufacturing
Water Level
1
1
1
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
0
0
0
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
Tickwise 1
Tickwise 2
Exchange Rate
1
1
1
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
0
0
0
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
Radio Waves
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
E*2
1
E*2
2
E*2
3
E*2
4
E*2
5
E*2
6
0
0
1
2
3
4
5
6
1
2
3
4
5
6
Fig. 6. A comparison of the SWAB algorithm with pure (batch) Bottom-Up and classic
Sliding Windows, on ten diverse datasets, over a range in parameters. Each experimental
result (i.e. a triplet of histogram bars) is normalized by dividing by the performance of
the worst algorithm on that experiment.
In addition, we have introduced SWAB, a new online algorithm, which
scales linearly with the size of the dataset, requires only constant space and
produces high quality approximations of the data.
There are several directions in which this work could be expanded.
•
The performance of Bottom-Up is particularly surprising given that it
explores a smaller space of representations. Because the initialization
phase of the algorithm begins with all line segments having length two,
all merged segments will also have even lengths. In contrast the two
other algorithms allow segments to have odd or even lengths. It would be
