[AstroPy] Python spectral fitting package - ala Sherpa

Andrew Ptak ptak at pha.jhu.edu
Wed Feb 4 10:00:24 CST 2004


Hello,

Sorry if this is getting a bit off-topic, but concerning these two, first 
the latest verison of XSPEC is object-oriented (written in C++) but hasn't 
been released yet.  This opens the possibility a little more of wrapping 
higher-level objects in python.  Second, I have a program that does 2-d 
fitting on images using Numeric (see http://xassist.pha.jhu.edu/ximgfit).  
It works fine for my purposes however I suspect it is a bit inefficient at 
the moment due to the creation of temporaries.  For example, since 
everything is in python the models are simply .py files (this of course 
was one of the design goals, to allow models to be added without 
recompiling).  Here is the code for computing a 2-d Gaussian model: 
     def Eval(self, x, y):
        xc = self.params[0].value
        yc = self.params[1].value
        norm = self.params[2].value
        sigmax = self.params[3].value
        sigmay = self.params[4].value
        theta = self.params[5].value
        cost = cos(theta/radeg)
        sint = sin(theta/radeg)
        dx = x-xc
        dy = y-yc
        rx = dx*cost + dy*sint
        ry = -dx*sint + dy*cost
        z2 = rx*rx/(sigmax*sigmax)+ry*ry/(sigmay*sigmay)
        # Avoid overflow errors
        z2 = self.min(z2, 100.)
        result = norm*exp(-z2/2)
        return(result)

In this design, x and y are 2-D arrays.  If I am not mistaken, in the line 
with z2 = rx* ... 7 temporaries are created.  From what I have heard 
weave.blitz() may be a solution but I haven't had a chance to play with 
that.  But my main point is that in general it may be tricky to be as 
efficient as the case where most of the computation is done at the C/C++ 
level since the programmer will have some insight, and with scripting 
languages only being used for the UI.

Andy Ptak

On Wed, 4 Feb 2004, Paul Barrett wrote:

> 
> 3. they only handle 1 dimensional response matrices (see e.g. FUSE,
>     which could use a 2D response matrix), and
> 
> 4. they are primarily implemented using compiled languages, so
>     development is slow.  (Sherpa's implementation is simply
>     bass-ackwards, i.e. imbeding an interpreted language in a compiled
>     language.  In addition, Numeric/numarray syntax is more powerful than
>     SLANG and hence, numerically more efficient.)
> 

_____________________________________________________
AstroPy mailing list  -  astropy at stsci.edu
http://lheawww.gsfc.nasa.gov/~bridgman/AstroPy/


More information about the AstroPy mailing list