[R] Fwd: Timebased predictions in postgresql.

Nicolai Petri nicolai at catpipe.net
Thu Aug 12 21:52:17 CEST 2004

Hi r-people :)

I'm sorry to disturb but I must admit that I know amazingly little about R and 
similar statistics-packages/languages and I'm kind of lost on where to start.
I'm currently working on a datacollection framework for postgresql (The db 
doesn't really matter except that I hope to use PL/R) and I would like to be 
able to predict future values preferable 1 day or more ahead. The highest 
resolution on the historic data is 4 minutes but I'm already resampling that 
to whatever I need, so if it would be better to use 30min or other 
reasolution (because of performance) it would be perfectly ok.

The types of statistics in the database is typically network io/ cpu usage, 
temperatur, etc. and I will rarely have holes in the history. 

Can anyone tell me how (or give me a hint) I can predict traffic or similar 
maybe one or more days ahead when I have the data xxdays back ? (And many how 
many days / which interval would be optimal for best performance/precision).

You can also tell me it's impossible, but I think it could be really cool to 
present graphs of expected cpu-load or network IO to our users.

Can it be done in R (and PL/R) ?

Best regards,
Nicolai Petri
catpipe Systems Aps

More information about the R-help mailing list