[SciPy-user] bug in cho_factor?

Emanuele Olivetti emanuele@relativita....
Sat Jul 19 16:25:27 CDT 2008


In the meanwhile I've submitted ticket #704 about this issue and
attached a tentative patch for the docstring.

I hope someone will review, improve it and commit.

Regards,

Emanuele

Warren Weckesser wrote:
> Hi Emanuele,
>
> On Sat, Jul 19, 2008 at 1:50 PM, Emanuele Olivetti
> <emanuele@relativita.com <mailto:emanuele@relativita.com>> wrote:
>
>     Thanks a lot for the explanation. I believe cho_factor's docstring
>     should be updated in order to mention these facts. It is definitely
>     unexpected that the result of the two decompositions are different
>     and can cause problems like I had (a couple of hours spent). A
>     clear "Warning" should fit. Consider that U,lower=cho_factor(A)
>     outputs an U that does not satisfy A==N.dot(U.T,U) !!
>
>
> I agree--the docstring description of the return matrix c is wrong.
> c is a matrix whose upper or lower (depending on the parameter
> lower) triangular part gives the Cholesky factor, but c itself is not
> triangular.
>  
>
>     Do you know why cho_factor does not zeros out the matrix?
>     Is it for performance reasons?
>
>
> That would be my guess, and it makes sense (why zero out elements
> that will be ignored by cho_solve()?), but you'd have to ask the
> author of the
> code to be sure.
>
> Warren
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> SciPy-user mailing list
> SciPy-user@scipy.org
> http://projects.scipy.org/mailman/listinfo/scipy-user
>   



More information about the SciPy-user mailing list