G'day davos,

I think I am okay from a fundamental clustering perspective - clustering 500 "cases" into 10 groups (clusters) based on 20,000 observations of each case. For smaller sets but the same sort of 'shape' I can perform this calculation in memory with no problems.

My key problem is that my single "fact" table, becomes 1 measure of TECK_VALUE.

I really need a measure of (TECK_VALUE when TECK_ID = 1), (TECK_VALUE when TECK_ID = 2)... one of these measures exists for each case. Whenever I run clustering now it defines each cluster in terms of TECK_VALUE --- this is no good, because I need to to consider how TECK_VALUE changes as TECK_ID changes... if that makes sense.