Fig. 4

Evaluation of S-RASTER (best viewed in color). As a shows, S-RASTER is faster than competing algorithms in terms of throughput. The algorithms processed 5M points in 10 batches of 500K points. The stated times refer to the end of each batch. As a sanity check, b plots the number of identified clusters. The input data set contained 100 dense clusters per batch. Windowed k-means was provided with the argument \(k = 100\). All data is the average of 10 runs