[SciPy-user] svdvals

[Pearu Peterson]

On Tue, 22 Oct 2002, Nils Wagner wrote:

> Hi Pearu,
> 
> Did you find anything explaining the strange behaviour of svdvals ?

No, I didn't.

Since it is very difficult to fix bugs that we cannot reproduce, I suggest
the following:
1) Try varying the input to svdvals to find out if there is any pattern
when svdvals returns nans. Try different arrays with different contents
and also with different dimensions, try also transposing arrays, etc.
2) Get lapack directly from netlib and use it for building scipy.
Also, try atlas from its homepage. See INSTALL.txt for instructions.

Pearu