Hi,

I'm having some trouble configuring a cube to achieve a hairy count
distinct problem. I have a fact table with three attributes per tuple:
a customer ID (primary key), a date, and a decimal x. So far I have
two dimensions, ID and date and measure x with aggregate function Sum.
This all works very well; Analysis Services makes it very easy.

The problem is that I can't work out how to count the number of
distinct customers who have an x > 0.0 such that the aggregates will
work over any of the dimensions and still return good data.

I have found a number of ways of doing it so far, none of which quite
work.

* I can make a view which adds a new column, xa, where xa = 1 iff x >
0.0. Then I can sum the xa column to find the number of customers but
this only works (because of the frequency of data collection) across
month aggregations and not years.

* Alternatively I can have a view which adds a new column, xb, where
xb = ID iff x > 0.0 and add xb as a new count distinct measure. This
has two drawbacks, first that the count distinct will be off by one
sometimes (as zero is a value in itself) and second that I have about
twenty columns like x in the actual fact table and having twenty count
distinct columns doesn't sound like a good idea even with virtual
cubes. (Using virtual cubes there is also the issue of having to share
dimensions if I want them to join properly.)

Can anyone suggest a good way of solving this problem? Much TIA.

--

Tom Rowlands
<mailto:tom (AT) ltc (DOT) act.edu.au.>
(Sorry, I can't sign this.)

A typical approach would be to define a fact table view that only
includes rows where x > 0.0, in which case the distinct count would not
be off by 1. But if you have 20 x columns, each with a separate
threshold, then this would entail 20 views and 20 physical cubes in a
virtual cube.


- Deepak

*** Sent via Developersdex http://www.developersdex.com ***
Don't just participate in USENET...get rewarded for it!