|
From: | alvarom |
Subject: | [Bug-gsl] Question about gsl_multifit_linear |
Date: | Wed, 28 Sep 2011 06:09:33 -0700 |
User-agent: | Internet Messaging Program (IMP) H3 (4.3.7) |
Hi,I?m using the function gsl_multifit_linear and I?m having the following problem. I?m using the function in a loop to approximate a function for different periods. The problem is that for different periods, I use different variables in the approximation. So what I do is: I define a matrix X with all the possible variables I?m going to use. The ones that I don?t need for a certain period, I define them as zero. Thus, when I run the OLS regression, the matrix X is not invertible. However, I thought this wasn?t a problem for gsl_multifit_linear. I though this function just simply eliminated the variables that create the problem of singularity (all the ones that are zero, given the regression includes a constant term). However, when I see the coefficients estimated by the function, some of those aren?t exactly zero. I checked the X matrix and it contains 0 where I intended to have 0, so I do not understand why for some coefficients I get a 0 value (as I expected), but for some others I get something close to zero, but not exactly 0 (for example, I get values like -2.19492e-08).
Thanks a lot, Alvaro.
[Prev in Thread] | Current Thread | [Next in Thread] |