Thread Subject:

MATLAB STATISTICS TOOLBOX SUCKS!!!

Maybe this will get some attention! So many distributions are missing and it can't even fit the data I posted on the forum in my thread "Noise from integer grayscale image". It sucks, sucks, sucks! dfittool is lame! I can't even fit a burr distribution or a pareto 2 (lomax) distribution or a wakeby distribution because MATLAB STATISTICS TOOLBOX does NOT even have them! Lame, lame, lame! I can't believe this product costs $200 for an individual license!

MathWorks has a dedicated toolbox for Image Processing (Image Processing
Toolbox)
Its quite possible that this toolbox has functions that address the specific
problem you are working on.

With this said and done, one of the nice things about MATLAB is that you
have the option to write custom functions that extend the basic capabilities
of the toolbox.

There is a very good demo on the Statistics Toolbox product page titled
"Fitting Custom Univariate Distributions". This demo shows some examples of
using the Statistics Toolbox function mle to fit custom distributions to
univariate data. Using mle, you can compute maximum likelihood parameter
estimates, and estimate their precision, for many kinds of distributions
beyond those for which the Toolbox provides specific fitting functions.

The demo is available at

http://www.mathworks.com/products/statistics/demos.html?file=/products/demos/shipping/stats/customdist1demo.html

One potential concern: From the looks of things, you are doing image
processing.
I'm not sure whether you want to be using a univariate distribution to model
your errors.

regards,

Richard


"Eric Diaz" <eric.diaz@gmail.com> wrote in message
news:in3go5$o9l$1@fred.mathworks.com...
> Maybe this will get some attention! So many distributions are missing and
> it can't even fit the data I posted on the forum in my thread "Noise from
> integer grayscale image". It sucks, sucks, sucks! dfittool is lame! I
> can't even fit a burr distribution or a pareto 2 (lomax) distribution or a
> wakeby distribution because MATLAB STATISTICS TOOLBOX does NOT even have
> them! Lame, lame, lame! I can't believe this product costs $200 for an
> individual license!

Just another couple examples of how much the statistics toolbox sucks!
1) There is no MLE fitting of central or noncentral chi squared distributions built in
2) There is no MLE fitting of threshold parameter of generalized pareto distributions (you have to know it in advance!!! How the hell am I supposed to know that MATLAB?)
3) There is no pareto 2 distribution.
4) There is no chi distribution fitting at all
5) There is no non-standard normal, i.e., sigma ~=1, based chi squared or chi based distributions!

ALL OF THE ABOVE, except 4 & 5, are provided free of charge by NIST Dataplot software. And for the $200 for an individual license that MATLAB charges, they should be provided, instead of me trying to write my own mle pdf, cdf, loglikelihood functions with censoring, etc...so that I can fit my data.

AND as for the image processing toolbox supporting a problem that is essentially statistical in nature...NOT LIKELY. I would be happy to be proved wrong though.

On Apr 4, 12:53 pm, "Eric Diaz" <eric.d...@gmail.com> wrote:
> Just another couple examples of how much the statistics toolbox sucks!
> 1) There is no MLE fitting of central or noncentral chi squared distributions built in
> 2) There is no MLE fitting of threshold parameter of generalized pareto distributions (you have to know it in advance!!! How the hell am I supposed to know that MATLAB?)
> 3) There is no pareto 2 distribution.
> 4) There is no chi distribution fitting at all
> 5) There is no non-standard normal, i.e., sigma ~=1, based chi squared or chi based distributions!
>
> ALL OF THE ABOVE, except 4 & 5, are provided free of charge by NIST Dataplot software.  And for the $200 for an individual license that MATLAB charges, they should be provided, instead of me trying to write my own mle pdf, cdf, loglikelihood functions with censoring, etc...so that I can fit my data.
>
> AND as for the image processing toolbox supporting a problem that is essentially statistical in nature...NOT LIKELY. I would be happy to be proved wrong though.

I'm not sure what good you think your whining is going to achieve -
apart from giving you the opportunity to let off steam, and make the
rest of us want to puke.
But if you channeled this aggression into solving your problem, you
might get further.
I don't have the Stats Toolbox, but I have no problem finding the
stats routines I need using a friend of mine called Google. All you
do is open Google and type into the little box something like "chi
distribution fitting Matlab" and Hey Presto!, you get hundreds of
links, and if you sort through those, almost invariably you'll find
exactly what you want.
As for complaining about the Stats Toolbox costing a whopping
$200!!!! You must be a student. For real people, it costs more than
ten times that.

LOL...Tideman you are so off base. As if I didn't spend a whole crap load of my time looking for solutions that MATLAB doesn't provide. To be honest, I spend way too much of my time doing that, looking for solutions that are not exactly what I need or sub-optimal. For the price paid for MATLAB toolboxes and the product itself, there is definitely too much of that going on. If no one speaks up about the poverty and/or mediocrity of content in MATLAB toolboxes, nothing will get done to improve them. Remember, MATLAB is a for profit company, and too often make tiny, incremental changes in their product, yet charge an exorbitant amount for an update.

On Apr 3, 11:16 pm, "Eric Diaz" <eric.d...@gmail.com> wrote:
> If no one speaks up about the poverty and/or mediocrity of content in MATLAB toolboxes, nothing will get done to improve them.  


But you're speaking up to the wrong people. Because . . . we don't
care! We don't care about you or your problem. When Image Analyst
told you (in the other thread) to speak up to the Mathworks about it,
you refused. Well you know what - they're the only people would would
care about your opinion of that toolbox. And if you won't explain
your problem to, apparently, the only people in the world who care
about your problem, then, well . . . you're on your own.

On Apr 4, 3:16 pm, "Eric Diaz" <eric.d...@gmail.com> wrote:
> LOL...Tideman you are so off base.  As if I didn't spend a whole crap load of my time looking for solutions that MATLAB doesn't provide.  To be honest, I spend way too much of my time doing that, looking for solutions that are not exactly what I need or sub-optimal.  For the price paid for MATLAB toolboxes and the product itself, there is definitely too much of that going on.  If no one speaks up about the poverty and/or mediocrity of content in MATLAB toolboxes, nothing will get done to improve them.  Remember, MATLAB is a for profit company, and too often make tiny, incremental changes in their product, yet charge an exorbitant amount for an update.

Actually, I agree with you.
I gave up maintenance back in 2006 because when I loaded 2006b, half
my programs that ran in 2006a no longer worked. TMW had arbitrarily
tweaked the system for no good reason that I could see. Why should I
pay big bucks for changes that don't benefit me? And will I have to
change dozens of .m files everytime a new version is introduced?
So, my Matlab is 5 years old now and still works perfectly.
I keep an eye on this forum, but I haven't seen anything that has been
introduced in 5 years that would justify upgrading from 2006a.