[SciPy-user] Multi-peak fitting
Thu Sep 27 19:44:23 CDT 2007
Are you using a point and click method of choosing initial parameters? I've
used that in the past and have code in matlab that does that.
Alternatively, for the case where I know the number of peaks, I use the
first and second derivative to look for initial parameters before using a
least squares method for the actual fitting. I've also implemented this in
MATLAB (still needs to be converted to python). For well separated peaks,
it works rather nicely. Now, I'm interested in trying something using
Bayesian methods to try to also determine how many peaks are present and
their positions in the case where I don't know the widths--I'd like to be
able to safely automate this procedure for when there are larger numbers of
files to be fitted and the point and click method becomes tedious, but where
the peaks are sufficiently overlapped that my current automated method gives
poor initial predictions. This would be along the lines of Sivia et. al.
Has anyone attempted an implementation of this yet?
(btw. for my current automated procedure, I haven't converted it over to
python yet because I didn't notice a savitsky-golay filter already
implemented in scipy--is there one now?)
On 9/27/07, Emmanuel <firstname.lastname@example.org> wrote:
> I think I may be quite interested in this kind of thing, if you want to
> share with me too.
> Emmanuel Favre-Nicolin
> On 9/27/07, Rob Clewley < email@example.com> wrote:
> > Brian, I'm just in the middle of developing some classes for feature
> > extraction in data, particularly for peaks in the kind of data you
> > have. The same classes store quantitative information (like peak
> > positions, heights...) that can be extracted by objective functions in
> > a fitting algorithm. The code will become part of PyDSTool (and as
> > such is somewhat integrated with scipy and numpy) but I am happy to
> > share with you what I have so far. It works well for the fitting
> > problems I am working on in my research and it's relatively
> > easy to use. It hasn't been tried on a wider range of problems yet. I
> > warn you that it comes with a little extra baggage because I'm setting
> > this whole thing up as part of a "bigger idea" involving model
> > estimation for dynamical systems and scientific data. But email me if
> > you want to try it out!
> > -Rob
> > _______________________________________________
> > SciPy-user mailing list
> > SciPyfirstname.lastname@example.org
> > http://projects.scipy.org/mailman/listinfo/scipy-user
> SciPy-user mailing list
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the SciPy-user