[SciPy-user] Re: SciPy-- convincing my lab to switch to SciPy/Python

Nils Wagner nwagner at mecha.uni-stuttgart.de
Wed Apr 27 14:05:17 CDT 2005


On Wed, 27 Apr 2005 12:51:02 -0600
  Travis Oliphant <oliphant at ee.byu.edu> wrote:
> Josh Tasman wrote:
> 
>> Python seems like a good choice for our lab, as it's 
>>relatively easy 
>> for  novice programmers to work with (vs C++ or perl.) 
>> More 
>> importantly, it has a large community behind it.
>>
>> Specifically, I need to convince my lab (and myself) 
>>that Python/SciPy 
>> is a robust replacement for Matlab.  Primarily, I need 
>>more info 
>> regarding the reliability of numeric computation. 
>> Programs like 
>> Matlab and Mathematica go through rigorous mathematical 
>>testing.  
>> What's the state of 
> 
>> SciPy in this regard?  Would you stand behind it for 
>> publication-quality work?
> 
> 
> The reliability of SciPy is somewhat mixed.     Most of 
>the actual algorithms are based on FORTRAN code that has 
>been around for literally decades and are quite reliable. 
>  On the other hand, you may still find bugs in some of 
>the less well-tested and not-as-often-used routines. 
>  Most of the time, these bugs are small easy-to-fix 
>Python-interface bugs.
> 
> Many people use and rely on the numerical algorithms in 
>scipy everyday.  Enthought uses scipy to deliver 
>numerical products to their clients.  Several researchers 
>use scipy in their publishable work.   Many scientists at 
>the national labs (Argonne, Livermore, Sandia, etc.) all 
>use Python for numerical work.
> 
> I would absolutely stand by SciPy for 
>publication-quality work.  I've published using SciPy. 
>  I think it's preferrable to proprietary solutions for 
>publications, because it actually allows someone to 
>reproduce your work and even inspect the code used to 
>produce your work.   So it is more in keeping with the 
>academic philosophy.
> 
>>
>> How would you compare SciPy's signal processing tools 
>>vs. Matlab's?
>>
> SciPy has quite a few of the tools that Matlab has (but 
>not everything, yet), and adding your own is not 
>difficult.   I've found it much easier, for example, to 
>add a tool to scipy then to go back to MATLAB for data 
>processing.
> 
> Wavelets are missing (but there is interest here and 
>some people are working on it).
> 
> Look at the docs at http://www.scipy.org/livedocs/  for 
>an overview.
> 
> If there is something you need that is missing, speak 
>up, maybe somebody can contribute it.
> 
Support for matrix equations (Sylvester, Riccati, Stein,
Lyapunov etc) would be nice.

Also sparse eigensolvers might be useful.

Nils


> -Travis O.
> 
> 
> P.S.
> 
> Scipy-Numeric is going through some changes that should 
>be in place in a couple of months.   My suggestion is to 
>start using scipy and then make the push to your people 
>in a few months.
> 
> 
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.net
> http://www.scipy.net/mailman/listinfo/scipy-user

  



More information about the SciPy-user mailing list