[Numpy-tickets] [NumPy] #923: numpy.random.logseries - incorrect convergence for k=1, k=2

[NumPy]

#923: numpy.random.logseries - incorrect convergence for k=1, k=2
--------------------------+-------------------------------------------------
 Reporter:  josefpktd     |        Owner:  somebody
     Type:  defect        |       Status:  new     
 Priority:  normal        |    Milestone:          
Component:  numpy.random  |      Version:  none    
 Severity:  normal        |   Resolution:          
 Keywords:                |  
--------------------------+-------------------------------------------------
Comment (by josefpktd):

 When comparing the c implementation in
 http://scipy.org/scipy/numpy/browser/trunk/numpy/random/mtrand/distributions.c
 with the python implementation from
 http://pal.ece.iisc.ernet.in/~dhani/frohne/rv.py

 it looks like that the numpy implementation only considers the algorithm
 for the case  # p is >= 0.95, however not the algorithm for  p < 0.95.

 However, running the above frequency test script, I get the opposite
 result, good approximation for very low p, while with high p, the
 convergence is pretty bad, so maybe the cases are reversed, but this is
 just guessing based on my sampling script.

 with prob p=0.02, I get

 {{{
 check frequency of k=1 and k=2 at N =  1000000
 0.989877 0.009962
 pmf at k = 1 and k=2 with formula
 [ 0.99    0.0099]
 }}}

 with prob p=0.95, I get

 {{{
 check frequency of k=1 and k=2 at N =  1000000
 0.200349 0.267545
 pmf at k = 1 and k=2 with formula
 [ 0.3171  0.1506]
 }}}

 Josef

-- 
Ticket URL: <http://scipy.org/scipy/numpy/ticket/923#comment:1>
NumPy <http://projects.scipy.org/scipy/numpy>
The fundamental package needed for scientific computing with Python.