Size of the Cache

We also did some experiments to measure the effect of the size of the cache on the training time. TABLE 5  shows the results for different cache sizes, from 10 to 100Mb. In these experiments, we used the first 10000 examples of each dataset (6192 for the smaller Kin) and used the non-sparse format. The only clear conclusion from these experiments is that the higher the size of the cache, the faster SVMTorch is, but the relation is completely problem dependent.

 

Table 5: Training time (in seconds) with respect to the size of the cache (in Mb).

	Size of the cache (in Mb)
	10	20	30	40	50	60	70	80	90	100
Kin	11	11	11	11	11	11	11	11	11	11
Artificial	359	182	106	100	99	98	98	98	98	98
Forest	869	715	622	519	450	391	355	316	305	272
Sunspots	7	7	7	7	7	7	7	7	7	7
MNIST	729	724	716	704	686	665	647	621	597	573