[R] One rather theoretical question about fitting algorithm

Bill.Venables at csiro.au Bill.Venables at csiro.au
Sun Jun 7 15:11:21 CEST 2009


The glm function allows you to specify starting values for beta (start =), eta (etastart =) or mu (mustart =).  In your case the starting value for beta would seem to be the most appropriate.  You might find providing a starting value for mu or eta also works well, provided you can come up with a quick way of finding values for the new observations.


________________________________________
From: r-help-bounces at r-project.org [r-help-bounces at r-project.org] On Behalf Of useR [milicic.marko at gmail.com]
Sent: 07 June 2009 21:20
To: r-help at r-project.org
Subject: [R] One rather theoretical question about fitting algorithm

Hi,

What I'm trying to achieve is very fast algorithm for fitting logistic
regression model. I have to estimate regression coeficients using
about 10k observations. Once I have coefficients estimated, new 100
rows of data becomes available.... Now I need to reestimate
coeficients using 100 newly arrived observations and removing 100
oldest observations.

So, my question is would it be possible to somehow reuse pre iusly
calculated coeficients and only adjust them cor newly arrived data? I
know it would have to be some aproximation but I suppose it will be
good enough.

I dont mind doing this in straight C because of of speed perative.
Actualy this will have to be cAlculated in a fraction of second.


Any ideas would be higly appreciated

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.




More information about the R-help mailing list