[Numpy-discussion] eigenvectors() hangs on nan's

[Norbert Nemec]

The current SVN version of numpy.linalg does not contain "eigenvectors"
any more (since the cleanup - patches I submitted last week), so you are
obviously using something older.

In any case, numpy.linalg.eig (and eigvals) do not seem to contain any
check against NaNs, which probably is a bad idea. scipy.linalg.eig does
contain such a check, but that module is internally very different to
that in numpy.

numpy.linalg certainly needs more cleanup. In that course, adding
safety-checks certainly is one task.




Chris Fonnesbeck wrote:

> If linalg.eigenvectors is called on a matrix that contains nan values,
> it hangs, quickly consuming system resources. Here is the matrix:
>
> matrix = [[ 0.          0.          0.          0.         
> 0.36658624  0.          0.
>    0.          0.          0.        ]
>  [ 0.89827259  0.          0.          0.          0.         
> 0.          0.
>    0.          0.          0.        ]
>  [ 0.          0.92510948  0.          0.          0.         
> 0.          0.
>    0.          0.          0.        ]
>  [ 0.          0.                 nan         nan  0.         
> 0.          0.
>    0.          0.          0.        ]
>  [ 0.          0.                 nan         nan  0.                 nan
>    0.          0.          0.          0.        ]
>  [ 0.          0.          0.          0.          0.93319244         nan
>    0.          0.          0.          0.        ]
>  [ 0.          0.          0.          0.          0.36658624 
> 0.          0.
>    0.          0.          0.        ]
>  [ 0.          0.          0.          0.          0.          0.
>    0.89827259  0.          0.          0.        ]
>  [ 0.          0.          0.          0.          0.         
> 0.          0.
>    0.92510948  0.          0.        ]
>  [ 0.          0.          0.          0.          0.         
> 0.          0.
>    0.          0.92510948  0.93319244]]
>
> I would expect an exception, but I get none.
>
> Using a recent cvs build of numpy.
>
> C.
>
> -- 
> Chris Fonnesbeck + Atlanta, GA + http://trichech.us