Fig. 2 depicts the cputime consumed in finding the h
leading eigenvectors for both the methods as a function
of data dimensionality. The data dimensions from 100 to
1000 are illustrated in Fig. 2 and dimensions from 2000
to 4000 are illustrated in Table 3. It is evident from
Fig. 2 that cputime curve for EVD based PCA increases
exponentially as the data dimensionality is increased.
Moreover, from Table 3 cputime for dimensions above
2000 are very expensive for EVD based PCA method and
may restrict practical applications which involve such high
dimensions. For dimension 4000 the EVD based PCA
method is consuming around 1413 cputime in seconds to
find 10 leading eigenvectors. On the other hand, it can be
observed from Fig. 2 as well as from Table 3 that the cputime
for Fast PCA method is very economical for all the
dimensions. Even for data dimensionality 4000, cputime
is only 7.53 s, which is extremely low as compared to
EVD based PCA method. Thus Fast PCA can also be efficiently
used for high dimensional applications.