On Sat, 18 Oct 2014, Rolf Turner wrote:
It would be a bit of work, although not an overwhelming intellectual
challenge, to produce an R package that would do essentially the same thing
as "easyfit". There are a number of questions that would have to be
addressed of course. E.g. just how do you want/expect the distributions to
be fitted to the data? Maximum likelihood? Are all the distributions dealt
with by "easyfit" amenable to being fitted via maximum likelihood? And how
is the choice of distribution to be made?
AIC? The "easyfit" web page refers to "goodness of fit tests", which can be
problematic, or "visual inspection" --- always a good idea, but it too can be
problematic.
Overall I don't think this "press a button and let the software do your
thinking for you" is the right way to go. If the results matter at all, you
need to know what you are doing and what pitfalls can lurk to trap the
unwary.
I don't understand what you mean by "All the R packages I've seen make you
build your own library of probability density functions and then do the
fitting on each one." R has a large number of built-in probability density
functions (including *most* of the distributions listed on the "easyfit" web
page) and most of these can be fitted (via maximum likelihood) using the
fitdistr() function from the MASS package. The fitdistr() function can fit
essentially any distribution for which a probability density function can be
written. Goodness of fit testing is more problematic, but then as I said
that is a problematic topic.
Superimposing fitted pdf-s on a histogram of the data for "visual comparison"
is straightforward.
cheers,
Rolf Turner
What I mean is that R has the capability of generating PDFs, and R has
the capability of calculating various goodness of fit measures, but if
you want to check goodness of fit measures against, say, 50 PDFs, then
you have to write the package. It's easier for me to use easyfit than
write the package.
This really isn't an issue of "press a button and let the software do
your thinking," it's more like "press a button and calculate 200 distance
measures." There's no particular virtue in doing the same thing 50
times, one by one.
billo
--
users mailing list
users@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe or change subscription options:
https://admin.fedoraproject.org/mailman/listinfo/users
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
Guidelines: http://fedoraproject.org/wiki/Mailing_list_guidelines
Have a question? Ask away: http://ask.fedoraproject.org
