R help - Apr 2009 - R-Squared with biglm?

I've been working with a rather large data set (~10M rows), and while biglm
works beautifully for generating coefficients, it does not report an r-squared.
It does report RSS. Any idea on how one could coax an R-squared out of biglm?

Thanks in advance for any help with this!

Bryan Lim
Lecturer
Department of Finance
University of Melbourne


      
	[[alternative HTML version deleted]]

On Mon, 20 Apr 2009, Bryan Lim wrote:
> I've been working with a rather large data set (~10M rows), and while
biglm
> works beautifully for generating coefficients, it does not report an 
>r-squared. It does report RSS. Any idea on how one could coax an R-squared
out
>of biglm?
Hmm. I don't ever use r-squared, so I didn't implement it. I'll add
the null model RSS to the output for the next verson

The obvious work-around is to fit a model with only an intercept, and divide the
RSS for the two models, but that takes twice as long.

The necessary information is all in the QR decomposition, the problem is just
getting it out. A Horrible Hack is to use the leaps package and do
   leaps:::leaps.from.biglm(model)$nullrss
to get the null model RSS.

    -thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle