Database Reference
In-Depth Information
Fig. 16. Average number of correct clusterings (out of 10) for 15 runs of the algorithm
using different sample sizes and
δ
values.
Fig. 17. Three recordings of the word
'norway'
in the Australian Sign Language. The
graph depicts the
x
position of the writer's hand.
furthermore the data seem to be already normalized and rescaled within
the range [
−
1
,...,
1]. Therefore in this experiment we used also the sim-
ilarity function
1 (no translation), since the translations were not going
to achieve any further improvement (Table 4, Figure 18). Sampling is only
performed down to 75% of the series length (these sequences are already
short). As a consequence, even though we don't gain much in accuracy, our
execution time is comparable to the Euclidean (without performing any
translations).
S
