I am working on some applications of using linear regression. Just wanted to check around here to see if I can download the CUDA based linear regression code somewhere?

Thanks a lot for the input!

I am working on some applications of using linear regression. Just wanted to check around here to see if I can download the CUDA based linear regression code somewhere?

Thanks a lot for the input!

GELS is Linear Solve. This is completely different from Linear Regression - which attempts to fit a curve using linear function of causal variables.

I am also interested in a CUDA version of Linear Regerssion (like Least Squarese tc…)

Are there libraries out there that can help?

Thanks,

To the above - linear regression is formed with the equation

Betas = (X’*X)(X’*y) which is precisely solving

(X’*X)betas = (X’*y), which is a linear solve.

What exactly does the use case look like? There are a variety of options when doing least squares computation. You may be able to use the normal equations or the pseudo inverse, or require a QR factorization (in which case there are multiple choices of how to implement QR decomposition).

What are the dimensions of the matrix (or matrices), how many matrices are there, and how many right-hand sides per matrix? How ill-conditioned are these systems? Is this a plain least squares computation, or least squares with constraints, e.g. NNLS?