[Numpy-discussion] Linear least squares

[Charles R Harris]

On Tue, Jan 8, 2013 at 11:17 AM, Till Stensitz <mail.till at gmx.de> wrote:

> Hi,
> i did some profiling and testing of my data-fitting code.
> One of its core parts is doing some linear least squares,
> until now i used np.linalg.lstsq. Most of time the size
> a is (250, 7) and of b is (250, 800).
>
> Today i compared it to using pinv manually,
> to my surprise, it is much faster. I taught,
> both are svd based? Too check another computer
> i also run my test on wakari:
>
> https://www.wakari.io/nb/tillsten/linear_least_squares
>
> Also using scipy.linalg instead of np.linalg is
> slower for both cases. My numpy and scipy
> are both from C. Gohlkes website. If my result
> is valid in general, maybe the lstsq function
> should be changed or a hint should be added
> to the documentation.
>
>
Do you know if both are using Atlas (MKL)? Numpy will compile a default
unoptimized version if there is no Atlas (or MKL). Also, lstsq is a direct
call to an LAPACK least squares function, so the underlying functions
themselves are probably different for lstsq and pinv.

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20130108/fe5220fc/attachment.html>