Table 2.

Random features required for ε and p

N	ε	p	ε	p
1	0.001	0.999	1 × 10-6	1.000
10	0.001	0.499	1 × 10-6	0.749
100	0.001	0.0667	1 × 10-6	0.129
1K	0.001	0.00688	1 × 10-6	0.0137
10K	0.001	0.000691	1 × 10-6	0.00138
>10K	0.001	≈ 6.91/N	1 × 10-6	≈ 13.8/N

The fraction of top feature sets, p, gets smaller as N increases. For p near zero, p ≈ -(ln ε)/N corresponding to the last column of the table.