Daniel Oberhoff wrote:
I can't do it as I have no access to the statistics toolbox, but
please
can someone else go through the distribution functions at least and
see
that the interfaces are consistent..
Hi, I think we have the toolbox here. I don't use Matlab much these
days since I don't have it on my laptop. How and what should I check?
Here is the help info for expcdf:
The functions that have already been checked are
exp{pdf,cdf,rnd,inv},
gam{pdf,cdf,rnd,inv}, and
norm{pdf,cdf,rnd,inv}
Frankly I think all of the other functions should be checked for
compatibility. The easiest is probably to go to the page
http://www.mathworks.com/access/helpdesk/help/toolbox/stats/
to identify all of the distribution functions.. I'd then check the
examples of at least the *cdf, *pdf and *inv functions for
agreement to
matlab's definitions. One way I could do this would be to just use the
examples on matlabs help pages for each of these functions and check
against Octave's results. An example is
http://www.mathworks.com/access/helpdesk/help/toolbox/stats/
index.html?/access/helpdesk/help/toolbox/stats/betapdf.html
However some help pages don't have examples and some might not
check all
possibilities of the distributions and so might not show a bug,
which is
why it would be better if some one with the statistics toolbox of
matlab
ran the tests. They could then test a larger number of possibilities
than those given in the help pages..
Basically we are just looking for different definitions and so the
errors are likely to be large. Things like the function using the
deviation rather than the variance...