Both data sets are reduced to two dimensions using dierent values of the
number of neighbors n. The dimensional reduction of the swissroll is evaluated
qualitatively, while the analysis of the tumor data set requires a statistical ap-
proach because of the complexity of the data. Specically for this purpose we
consider the percentage of nearest points in the original space that are preserved
as nearest neighbors in the dimensional reduced space, and the stress induced
by the dimensional reduction. In addition, in the nal part the running times of
LLE and ISOLLE are compared.