A Multiscale Clustering Approach for Non-IID Nominal Data

Algorithm 3 Downscaling algorithm based on Lanczos (DSAL).

 Input: list(Rcenter [1], Rcenter [1],…, Rcenter[n])//the clustering result of benchmark-scale dataset

 Output: the clustering result of target dataset

 1: design kernel function L(x) = sinc(x) sinc(x/a)

 2: for i in range(n)://n is the number of cluster belonged benchmark scale

 3:  for j in range(i+1,n):

 4:   Mij = getSimilarCMS(Rcenter[i], Rcenter[j]);

 5:   convert (Rcenter[j]) by using M and L(x).

 6: return Rcenter[]//the clustering result of target dataset