[SciPy-User] problem with discrete rvs?

[Robin]

Hi,

If I understand the scipy.stats package I should be able to create my
own arbitrary discrete distribution and generate samples from it... I
am trying this but am having the following problem. I define my
discrete distribution with the following:

vals = (array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13,
14, 15, 16,
       17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31]),
 array([ 0.31518555,  0.08191701,  0.0817977 ,  0.02212394,  0.08149974,
        0.02204335,  0.02201125,  0.00595341,  0.0817977 ,  0.02212394,
        0.02209172,  0.00597518,  0.02201125,  0.00595341,  0.00594474,
        0.00160788,  0.08161887,  0.02207557,  0.02204342,  0.00596212,
        0.02196313,  0.0059404 ,  0.00593175,  0.00160437,  0.02204342,
        0.00596212,  0.00595343,  0.00161023,  0.00593175,  0.00160437,
        0.00160203,  0.0004333 ]))

Then I do
rv = rv_discrete(name='test',values=vals)

and then try to generate samples. When I plot the results of this for
large number of samples, it seems to match pretty well but the last 4
or 5 values are never generated and have zero probability, even with
enough samples to really capture them.

In [742]: (te.rvs(size=1000000)==30).sum()
Out[742]: 0
In [743]: (te.rvs(size=1000000)==29).sum()
Out[743]: 0
In [744]: (te.rvs(size=1000000)==27).sum()
Out[744]: 0
In [745]: (te.rvs(size=1000000)==20).sum()
Out[745]: 21884

Does anyone have any ideas? Is this a bug in rv_discrete.rvs?

In [746]: scipy.__version__
Out[746]: '0.8.0.dev5825'
In [747]: numpy.__version__
Out[747]: '1.4.0.dev7039'

Cheers

Robin