I've been trying for a long time to figure out how to perform (on paper) the divisive hierarchical clustering algorithem, however I'm not able to understand how to do it exactly. example:
I need to do it using Manhattan distance.
According to what I understand, I need to calculate the distances matrix (using Manhattan distance). then put all the observation in one cluster (let's call it A) and have another cluster, which at the beginning will be empty, B.
Then I need to calculate: a(1)=[d1,2+d1,3+d1,4+d1,5]/4 for all i in 1,...,5 and calculate a(i)-d(i,B) and choose the observation the gives me the maximal a(i)-d(i,B) and move it to B.
Do I need to calculate again the distances matrix? How do I calculate d(i,B) when be contains more then one observation?