Top: histogram of coincidence for pairs of classifications, one inferred with k-means and the other with single pass k-means. The mean coincidence, around 70%, is characteristic of the SDSS-DR7 galaxy spectrum catalog (Sánchez Almeida et al. 2010). Bottom: same as avobe, except that only classifications using k-means are compared. It shows the intrinsic dispersion in possible classifications due to the random initialization of the algorithm.