0
$\begingroup$

I read it to be similar to PCA so confused on what metric it chooses to split. For eg. if for hierarchical varclus, does it calculate correlation matrix and then simply divide this into "x" bins/clusters? OR

Does it create "x" components. Then try to "regress" this component using combinations of the independent variables and find which explained the highest variance ?

@Anony-Mousse suggested taking transpose of the original matrix wich then reduces it to the common clusterng scenario where distances are calculated. Like in normal kmeans/hierar each axes belongs to each variable and thus each point is plotted in the space.

When we transpose but, what becomes of the axes? Like each point is represented in which manner in the space.

$\endgroup$
3
  • $\begingroup$ You probably mean clustering the transpose of a matrix? You can choose any distance measure there. You'd probably use the average of the clustered columns afterwards. $\endgroup$ Apr 15, 2018 at 23:44
  • $\begingroup$ @Anony-Mousse thats a good way of looking thru. In this case then, what would be the axes?Like in normal kmeans/hierar each axes belongs to each variable and thus each point is plotted in the space. When we transpose, what becomes of the axes? $\endgroup$ Apr 16, 2018 at 2:01
  • $\begingroup$ The data points. $\endgroup$ Apr 16, 2018 at 7:14

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.