[SciPy-user] pyloess in scipy?

[Matthew Brett]

Hi,

On Tue, Sep 23, 2008 at 10:03 AM, Pierre GM <pgmdevlist at gmail.com> wrote:
> On Tuesday 23 September 2008 12:45:45 Jarrod Millman wrote:
>> Here is pyloess:
>> http://projects.scipy.org/scipy/scipy/browser/branches/sandbox/scipy/sandbo
>>x/pyloess
>
> Jarrod,
> Thanks a lot. Would it be useful to clean up the package and make it a proper
> scikits ?

I'm attaching a slightly modified version of the pure python biopython
implementation.

However, I don't know whether it (or the original) works correctly.

This in hope that someone who knows this stuff better than me is interested.

Biopython implementation here:

http://cvs.biopython.org/cgi-bin/viewcvs/viewcvs.cgi/biopython/Bio/Statistics/lowess.py?cvsroot=biopython

I was expecting this to work:

n = 100
x = np.arange(n)
y = np.random.normal(size=(n,))
yest = lowess(x, y)

but it returns a LinAlgError (as does the unmodified original).

I see that R has a more complex algorithm, based on the original
fortran.  For reference, I have attached the file (lowess.doc) that
they worked from, that used to be in the R source.

Any thoughts?

Matthew
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lowess.doc
Type: application/msword
Size: 20671 bytes
Desc: not available
URL: <http://mail.scipy.org/pipermail/scipy-user/attachments/20080924/6f6d1883/attachment.doc>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lowess.py
Type: text/x-python
Size: 3444 bytes
Desc: not available
URL: <http://mail.scipy.org/pipermail/scipy-user/attachments/20080924/6f6d1883/attachment.py>