Statistics function incorrectly computing median

[Ben Abbott]

On Jan 14, 2008, at 7:17 PM, John W. Eaton wrote:

> On 13-Jan-2008, Ben Abbott wrote:
>
> | I took another look and found a method that is more consistent with
> | the original algorithm. The original calculates the CDF from left to
> | right.  So I added an equivalent calculation form right to left, and
> | averaged the two results.
> |
> | I expect the result will be the same calculated by GNU's R.
>
> Why not look at the R code for this function and see what it is doing?
>
> jwe

I looked at the routines in R and Maxima and have concluded that they  
assume the probability of each sample is a constant; i.e. 1/(# of  
samples).

The routine "discrete_inv (x, v, p)" permits the user the specify the  
probability of each sample.

Thus, it took some time to determine a method that was consistent with  
constant probability and variable probability.

Before, submitting a formal patch, I thought it wise to attach a  
script for testing.

The equivalent commands in Octave, R, and Maxima are

Octave: discrete_inv ([1/4 2/4 3/4], <v>, ones (size (<v>)) / numel  
(<v>))

GNU R: summary ( c( <v> ) )

Maxima: load( discriptive )$
         quantile ( <v>, 1/4), numer;
         quantile ( <v>, 2/4), numer;
         quantile ( <v>, 3/4), numer;

Where <v> is to be replaced by a sequence like; [1, 2, 3, 4].

I've compared the result for a few examples. It appears to now be  
working properly. However, I've been wrong before, so ....

Ben



-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://www.cae.wisc.edu/pipermail/bug-octave/attachments/20080118/44fe9a5c/attachment.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: discrete_inv.m
Type: application/octet-stream
Size: 2893 bytes
Desc: not available
Url : https://www.cae.wisc.edu/pipermail/bug-octave/attachments/20080118/44fe9a5c/attachment.obj 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://www.cae.wisc.edu/pipermail/bug-octave/attachments/20080118/44fe9a5c/attachment-0001.html