A graphical heuristic for reduction and partitioning of large datasets for scalable supervised training

Journal of Big Data

Table 14 Scaling improvements of the heuristic approaches

# of training points	Shrinking heuristic	GSH	serial GCH	dist. GCH	dist. GCH
	of LIBSVM	GSH	serial GCH	(w/2 proc.)	(w/4 proc.)
	(seconds)			(\(\times \) of serial GCH)
1000	0.0078	0.0076	0.0073	1.5	3.0
4543	0.0604	0.0339	0.0290	1.7	2.3
9686	0.1957	0.0981	0.0631	1.3	1.6
20,647	0.7422	0.2515	0.1359	1.8	2.2
93,823	9.6774	2.3658	1.0003	1.8	3.6
136,984	20.4771	3.9556	1.9835	1.5	2.0