Table 3.2 presents a summary of the experimental results
with the canopy clustering, where the threshold parameters
and the number of clusters are tuned on the separate validation
set. The canopy clustering algorithm achieves an F1
of 0.838 in only 7.65 minutes. In comparison, the complete
clustering takes over two hours, and has slightly worse error.
Note that this represents more than an order of magnitude
reduction in computation time.