Welcome to MathOverflow

Question

1

I have a probability mass function of some experimental data who's log looks like the following: (please ignore the fact that it is not normalized) alt text

(meaning if p(x) is the pmf, this is log(p(x)) ) Does anyone know what parametric family it might belong to? (note that this is a discrete distribution)

flag	asked Feb 9 2010 at 22:06 liza 181●7

1

The graph kind of looks like a rotated Fermi-Dirac distribution... Can you give some information about where this data is coming from? Do you have any theoretical model to compare it to? – Alberto García-Raboso Feb 9 2010 at 22:14

Sigmoids ( en.wikipedia.org/wiki/Sigmoid_function ) are all over the place in probability laws. You should try to fit the examples linked to in Wikipedia. – Steve Huntsman Feb 9 2010 at 22:21

1

if anything it is more of an inverse sigmoid. It came from the distribution of certain image features after they have been quantized (it is the distribution of a sort of 'visual vocabulary'). The distribution is from a large database of images with no special characteristic (natural + synthetic images). – liza Feb 9 2010 at 22:30

1

So the y axis is log-number of features falling in a particular bin, and X is just feature index sorted by popularity? – sheldon-cooper Feb 9 2010 at 22:44

sheldon: exactly – liza Feb 9 2010 at 22:47

Felipe Olmos · Answer 1 · 2010-02-22 14:31:38Z UTC

3

I tried taking the logarithm to the pdf of a $\beta(0.5, 1.5)$ (see Beta distribution) and it gave me this

beta

Maybe this can be fitted for your data.

answered Feb 22 2010 at 14:31

Felipe Olmos
152●8

2

The Beta-family seems a really good guess, though at this point it is just what it is: an educated guess. If liza could explain to us what and how has been measured, we might say more :). – fedja Feb 22 2010 at 15:59

Very interesting! I'll try it – liza Feb 23 2010 at 20:37

probability mass function fitting

Remember to vote up questions/answers you find interesting or helpful (requires 15 reputation points)

1 Answer

You can accept an answer to one of your own questions by clicking the check mark next to it. This awards 15 reputation points to the person who answered and 2 reputation points to you.