[R] machine precision problem

[David Winsemius]

On Feb 21, 2010, at 1:37 PM, wendy wrote:

>
> Dear all,
>
> I am analysising a set of gene expression data using multiplicative  
> Gaussian
> Kernel function. I calculated the kernels between two gene sets with  
> 20000
> genes for each set. The kernels are less than 1, therefore the  
> product of
> the kernels is out of the minimum precision in R. I wonder if there  
> is a way
> to lowering the minimum double precision in R. I am using R version  
> 2.10.1.
>

Have you considered other strategies, e.g. summing the logs and  
exponentiating?

Your proposed solution would involve rewriting R. There is no switch  
to throw that would increase the width of doubles.

-- 
David