basic statistics in python

[John Gorman]

Doug,
    Another consideration is using the MLab module and Numeric.  MLab 
emulates several Matlab function calls including "cov" (which doesn't 
have a docstring, by the way, but here's Matlab's definition of cov():
   """COV Covariance matrix.
          COV(X), if X is a vector, returns the variance.  For matrices,
          where each row is an observation, and each column a variable,
          COV(X) is the covariance matrix....
   """

    Anyway, you could write your own python-based code, but the cov() 
function is likely to be more efficient for large arrays.  Here's an 
example using the scipy module (which contains Numeric and MLab):

Python 2.1.1 (#20, Jul 20 2001, 01:19:29) [MSC 32 bit (Intel)] on win32
Type "copyright", "credits" or "license" for more information.
IDLE 0.8 -- press F1 for help
 >>> import scipy
 >>> # generate a vector of std nrml rv's
 >>> v = scipy.normal(0,1,100)
 >>> # cov returns the variance if its argument is a vector
 >>> # rather than a matrix
 >>> sig_v = scipy.sqrt(scipy.cov(v))
 >>> print sig_v
0.97295502908

    I haven't peeked under the hood with scipy.normal, but I'm assuming 
they're using something similar to the Box-Muller algorithm... Perhaps I 
ended up with a "bad seed" here (expecting sig_v = 1) ;^).  Hopefully 
you get the picture... ;^)

    Finally, if you want access to other statistical routines there are 
some python based modules floating around including Scientific Python 
and Gary Strangman has stats.py at: 
http://www.nmr.mgh.harvard.edu/Neural_Systems_Group/gary/python.html

Cheers,
John Gorman


Fernando Pérez wrote:

> dsavitsk wrote:
> 
> 
>>i found this
>>http://starship.python.net/crew/hinsen/ScientificPythonManual/
>>but it seems to be unix only and i am on win32. as far as standard
>>
> 
> you can probably get the stuff you want running under win. I'd be very 
> surprised if all of Konrad's code was tied to unix (though I can't tell for 
> sure, I use it under unix).
> 
> 
>>There seems to be some stuff in scipy, but i am having trouble geting
>>through the docs.  also, do i want to look at numpy?
>>
> 
> Scipy is early and still a bit rough to install, but fabulous. And yes, you 
> NEED numpy for anything numerical in python above the toy level. Both 
> konrad's libs and scipy are numpy based anyway.
> 
> cheers,
> 
> f
>