test of normal distribution functions

[Prof J C Nash]

Morten is right that some values input for inverses tests are not fair 
to any spreadsheet processor. I hope to find ways to provide more 
informative output, that is, how the calculation fares in relation to 
what is reasonable at the default precision. The current example is VERY 
preliminary, and I simply copied values from one test to another.


Some of the more interesting tests are the bounds ones. These allow 
comparison with simple functions that can be locally computed. There are 
other examples of such functions, and a compromise is needed between 
different objectives to get a nice test.


For the moment I am avoiding the r.foo functions. Later on I hope to add 
to the Gnumeric-specific test files. For the moment I'm concentrating on 
ones that can be saved as .xls. I find these useful to send to Excel 
addicts.


JN


--
John C. Nash, School of Management, University of Ottawa,
Vanier Hall 451, 136 Jean-Jacques Lussier Private,
P.O. Box 450, Stn A, Ottawa, Ontario, K1N 6N5 Canada
email: nashjc on mail server uottawa.ca, voice mail: 613 562 5800 X 4796
fax 613 562 5164,  Web URL = http://macnash.admin.uottawa.ca
Practical Forecasting for Managers web site is at
http://www.arnoldpublishers.com/support/nash/
___
gnumeric-list mailing list
gnumeric-list@gnome.org
http://mail.gnome.org/mailman/listinfo/gnumeric-list

Re: test of normal distribution functions

[Morten Welinder]

One more quite important thing:

You are testing things that should be equal and not
things that should be different.

Some spreadsheets like OO and Excel cheat in the - and
= operators: things that are mostly equal will be considered
equal, i.e., - will return exactly zero.

Such spreadsheets are more likely to pass should be equal
tests regardless of their precision!  On the other hand, they
will flunk should be different tests more easily.

As a thought experient, consider how many of your tests would
be passed if - always returned 0 and normsdist returned 42.
It had better not be all!

Morten
___
gnumeric-list mailing list
gnumeric-list@gnome.org
http://mail.gnome.org/mailman/listinfo/gnumeric-list