Fig. 3

Flow graph of S-RASTER, a modification of RASTER (cf. Fig. 1) for evolving data streams. Refer to the Sect. “Idea” for a detailed description. The input source node s distributes values arbitrarily to projection nodes \(\varvec{\pi }\), which reduce the precision of the input. In turn, they send projected values to accumulation nodes \(\varvec{\alpha }\). These nodes keep a count of points for each tile and determine significant tiles for the chosen sliding window. Should a tile become significant or a once significant tile no longer be significant, a corresponding update is sent to the clustering node \(\kappa\). Node \(\kappa\) periodically performs clustering of significant tiles, which is a very fast operation