[SciPy-user] Question about scipy.stats

[Jesús Carrete Montaña]

El Sábado, 5 de Mayo de 2007 22:14, Donald Fredkin escribió:
> Jesús Carrete Montaña wrote:
> >    I get the following error when trying to evaluate the
> > probability density function for a Poisson distribution:
> >
> > from scipy import *
> >
> > d2=stats.poisson(loc=1.,scale=2.)
> > x=arange(1,10,1)
> > d2.pdf(x)
>
> What is a probability density function (pdf) for a Poisson
> distribution? and what does scale mean? Both notions are fine for a
> Gaussian distribution, which is continuous, but the Poisson
> distribution is discrete and contains only one parameter.

While your observations are formally correct, the documentation for 
scipy.stats states that the pdf() method should exist - I suppose it 
should work as the discrete counterpart of a true PDF, returning the 
probability of its argument (or raising an exception if it's not an 
integer). Similarly, although the Poisson distribution is 
uniparametric, the same documentation says that those two arguments 
should be accepted by all the constructors (even if they are ignored). 
I included them only to make the parallelism between my calls to 
poisson() and norm() more evident.

To avoid confusion, let's take an example of a continous, biparametric 
distribution such as scipy.stats.beta:

XXXXXXXXXXXXXXXXXXXX--code--XXXXXXXXXXXXXXXXXXX

import scipy.stats
d=scipy.stats.beta(loc=.2,scale=.1)
d.pdf(.3)

---------------------------------------------------------------------------
exceptions.TypeError                                 Traceback (most 
recent call last)

/home/kuantiko/latex/2006-2007/lFNyP/memoria/<ipython console>

/usr/lib/python2.4/site-packages/scipy/stats/distributions.py in 
pdf(self, x)
    104         self.dist = dist
    105     def pdf(self,x):
--> 106         return self.dist.pdf(x,*self.args,**self.kwds)
    107     def cdf(self,x):
    108         return self.dist.cdf(x,*self.args,**self.kwds)

/usr/lib/python2.4/site-packages/scipy/stats/distributions.py in 
pdf(self, x, *args, **kwds)
    466         goodargs = argsreduce(cond, *((x,)+args+(scale,)))
    467         scale, goodargs = goodargs[-1], goodargs[:-1]
--> 468         place(output,cond,self._pdf(*goodargs) / scale)
    469         return output
    470

TypeError: _pdf() takes exactly 4 arguments (2 given)

XXXXXXXXXXXXXXXXXXXX--code--XXXXXXXXXXXXXXXXXXX

The exact same code, but changing 'beta' to 'norm', works as expected. I 
suppose I am doing something wrong (given the lack of answers to my 
original message, probably something very elementary) but I still 
haven't been able to find the solution. I'm using the functions from 
Rpy in the meantime: great software, but IMHO terribly slow. I would 
prefer to use only scipy and would e very grateful if someone pointed 
me to the answer.

Thanks,
Jesús.