[SciPy-dev] slow setitem

Gael Varoquaux gael.varoquaux@normalesup....
Wed Dec 2 02:41:18 CST 2009

On Wed, Dec 02, 2009 at 09:24:27AM +0100, Benny Malengier wrote:
> So, looking back at this, I think the main problem for scipy is:
> 1/need to add documentation in functions as setdiag and in base
> classes that it is slow, so not good to use in loops
> 2/need to add documentation in
> http://docs.scipy.org/doc/scipy/reference/sparse.html in the API of
> sparse matrixes, so indicating one can access data, indexes, .... of
> csc and csr matrixes to achieve great speedup (if you know what you
> are doing).
> 3/need to add documentation with example of how one can solve sparse
> matrixic using lapack. A new users that wants to solve a sparse matrix
> system with scipy looks in the documentation and arrives at spsolve,
> unaware it probably is not the best way to solve his problem. It is
> strange for experienced people to see new PhD students starting and
> having no idea that blas and lapack exist.
> Note that if you search in google on 'lapack scipy' you do not find
> any example at all.

Hey Benny,

If you get time, you should know that you can easily improve the scipy

 1) Read http://docs.scipy.org/numpy/Front%20Page/
 2) Get a login
 3) On the page: http://docs.scipy.org/doc/scipy/reference/sparse.html
    (or any other documentation page) click on the 'edit page' link, 
    on the bottom left.

As you have experienced, there is a welth of knowledge to acquire to be
efficient with numerical calculation. Presenting it to the user in a way
that he finds it quickly without being drowned by needless information is
hard. Input from a user that has just gone through the process of
learning is invaluable.


More information about the SciPy-Dev mailing list