I'm puzzled with normfit.
I have intensity of velocity data on a given X-direction, converted as a 1-D vector Vx. It has a size of S= 256 X 256 elements (65536) elements. They are signed values, of the order of 10^-2.
No problem with dfittool. I get sensible values for mean and standard deviation. But i need to automate the process, for batch processing. No GUI. Thus, thought i used normfit to fit a PDF. 1) Is this the right way to go ? Is normfit designed to fit a PDF with signed bins values ?
I thought i'd output the histogram data and built the PDF from there : So i did :
Then to obtain pdf i did :
pdf = nVx ./ (S*diff(Vxbins(1:2)))
A posteriori, i checked that actually sum(nVx) = S = 65336 (makes sense to me)
From there, not sure i should normalize to the max... does normfit require that ? My max value in the pdf is about 15.
From there, i do
[mu , sigma, blah...] = normfit(pdf)
Result is crap, i get something completely different from dfittool.
What am i missing ? Is normfit not designed for this situation ? If not, what is the right way to go ?
No products are associated with this question.
Probably the easiest way to handle this, if you like what the distribution fitting tool is doing, is to autogenerate code from the tool itself. You can do that via File -> Generate Code. That will allow you to do batch processing.
You should be calling normfit on the raw data vector itself, not your pdf approx:
[mu , sigma, blah...] = normfit(Vx)
Thank you for your answer. I just tried generating the code. I found out that they are using fitdist in a way i didn't use before. They are not using directly normfit. So i assume normfit is more low level and assumes X to be positive (while my data are signed), although the help doesn't say anything about that. So, using fitdist is much easier and answers my question, although the use of normfit stays really obscure. (concerning the domain for X...)