RE: Grouping My Graphs

This is called piecewise multiple linear regression. A net search will find you some references to using data mining in this regard. In your case time is the relevant smoothing spline.

Instead of trying to co-ordinate your graphs, you might do well to toss the various variables versus time into a program that reports the correlations together with confidence calculations. That will save you a lot of analysis on things that “look” like a correlation but have no statistical relevance. Even statistically relevant correlations may have no causal relationship, but at least there is a reasonable chance they might.

Regards,

mwf

-----Original Message-----
From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org]On Behalf Of Ethan Post
Sent: Friday, April 21, 2006 2:47 AM
To: _oracle_L_list
Subject: Grouping My Graphs

Here is a problem. I have 300 nice pretty graphs. Some graphs are flat, some show spikes and valleys, some show slow growth and slow decline. What would be really nice is to find some easy way of grouping charts that likely have correlation of some sort.

For example, if I have a chart with a peak at 1am and 4am, it would be grouped with other charts that show a spike or valley at 1am and 4am, since they likely correlate.

I am sure there must be some way to do this in PLSQL or SQL and I just thought of this so I have not pondered a solution yet. Has anyone ever heard of anything like this? My guess is I would profile the data for a chart, then go look at all the other charts for the same "profile" and assign the chart to a group within the current profile. A profiling the spikes and valleys would be easy, but slow declines and increases, aka hills, would be much harder to "see" pragmatically.

Perhaps this is one for Joe Celko.

• Ethan

--
http://www.freelists.org/webpage/oracle-l