How to keep a certain level of the clusters from de Proc Cluster

Mathis1 — Mon, 18 May 2020 10:24:21 GMT

Hi,

I ran a proc cluster on the proc corresp output. I obtain a tree (see picture below), and the output table (see table attached).

In the table (Ward), I have in front of each of my variables, the name of (what I believe to be) the smallest cluster I can get. What i'd like, one of my most cherished dreams, is having several columns in my Ward table, each column representing one level of clustering.
For instance Var_Level_1 : Cluster_1 ; Cluster_2

Var_Level_2 : Cluster_1_1 ; Cluster_1_2 ; Cluster_2_1 ; Cluster_2_2 etc...

Thank you by advance for your help 🙂

Re: How to keep a certain level of the clusters from de Proc Cluster

RichardDeVen — Wed, 20 May 2020 10:00:15 GMT

A set of _NAME_ _PARENT_ data nodes define a hierarchical tree. From node X, following the data up the parent linkage, to the root is an ancestors path. You want data columns , 1..N, containing the names from root to X. The data node can be stored in a DATA step HASH object and ancestral path traversal can be performed using a series of FIND() operations.

The number of steps it takes to traverse from a given node to the root is the depth of the node.
Suppose you have some longest path P > Q > R > S > T > U > V > W which is length depth 7.
- You know you will need to have 7 columns to capture the pieces of the longest path.
- This determination (computation) needs to be done before a final DATA step.
Suppose you have another path A > B > C > D of depth 3.
- Traversing up the parent links from D you have
  - step 1 C
  - step 2 B
  - step 3 A
  - step 4 done
- Arrayed, the traversal is in reverse order C B A, you want the data as A B C.
  - The data captured during traversal needs to be reversed

Example:

* pass 1 - compute number of columns needed; * find longest path; data _null_; if 0 then set download.ward; * prep pdv; declare hash links (dataset:'download.ward'); links.defineKey('_NAME_'); links.defineData('_PARENT_'); links.defineDone(); declare hiter iter('links'); do index=1 by 1 while (iter.next() = 0); do step = 1 by 1 while (links.find(key:_parent_) eq 0); if _parent_ = '' then leave; end; max_step = max (max_step, step); end; call symput ('MAX_DEPTH', cats(max_step)); run; %put NOTE: &=MAX_DEPTH; * final step; data have; length depth 8; length level1-level&MAX_DEPTH $25; array level level1-level&MAX_DEPTH; set download.ward; if _n_ = 1 then do; declare hash links (dataset:'download.ward'); links.defineKey('_NAME_'); links.defineData('_PARENT_'); links.defineDone(); end; * determine depth (number of steps to root) and capture tiers; * in this loop 'level' actually means 'ancestor'; orig_parent = _parent_; level(1) = _parent_; do step = 2 by 1 while (links.find(key:_parent_) eq 0); if _parent_ = '' then leave; level(step) = _parent_; end; _parent_ = orig_parent; depth = step - 1; * reverse the the captured tiers; do step = 1 to depth/2; opstep = depth - step + 1; hold = level(step); level(step) = level(opstep); level(opstep) = hold; end; drop step opstep hold orig_parent; run;

topic How to keep a certain level of the clusters from de Proc Cluster in Statistical Procedures

How to keep a certain level of the clusters from de Proc Cluster

Re: How to keep a certain level of the clusters from de Proc Cluster