[R] Updating a linear model

[Paul Meagher]

Say I have collected data and used it to construct a linear model.

I now have a new observation and want to use it to update my linear model.

Is there a more efficient way to update the model than recomputing the
linear model from the complete data set?

In other words, can I incrementally update a linear model as new
observations come in without recomputing the linear model from the full data
set?   I've looked at the help for the lm package but can't see anything
obvious that answers my question.

I am trying to use a set of regressors to predict what the next observation
will be.  I would like my coeeficient and intercept estimates to improve as
the data comes in - it will need to work well under limited information at
first (or not at all if there is some lower bound of information quantity
that is required).  If the next data point comes in, I will have (binary or
quantitative) information about whether the prediction was successful or
not.  Not sure if the option to use feedback would dictate a different
approach than the one I am thinking of.   Not sure if a backprop neural
network can be used to estimate the coefficients in a dynamically evolving
multiple linear regression equation.

If anyone has any pointers to packages I might want to look at, suggestions,
etc... I would appreciate it.

Regards,

Paul Meagher

Datavore Productions
50 Wood Grove Drive
Truro, Nova Scotia
B2N-6Y4
1.902.895.9671
www.datavore.com