Suppose you have a variety Inventory Items, each with their only inventory levels and you have some historic sales for each Item type. If you want to group your inventory items into clus tersthat behave similarly (w.r.t sales), what is the optimal way? It doesn't make much sense to cluster by absolute sales, since inventory levels vary across item types. It also doesnt make much sense to cluster by sales/stock ratio (or the log of that) for the same reason.

8/6/2012 3:55:07 PM

What's a "tersthat" and what are "clus"?

8/6/2012 8:01:44 PM

how much data are you talking about

[Edited on August 6, 2012 at 9:16 PM. Reason : you could weight the clustering between absolute sales and inventory turns]

8/6/2012 9:14:11 PM

^ Millions of rows of sparse data.

Im thinking something like log(sales*sales/stock) bucketed into weeks.

8/7/2012 9:31:50 AM