Dear Ken, Routines to fit a Zipf-Mandelbrot (probability) distribution are implemented in our zipfR library for R:
http://www.cogsci.uni-osnabrueck.de/~severt/zipfR The library documentation also points to some literature on the topic (most notably, Baayens' 2001 Word Frequency Distributions book). Regards, Marco Ken Williams wrote: > Hi, > > I've got some data (picture at http://limnus.com/~ken/zipf.jpg ) that seems > to very obviously follow a Zipf-Mandelbrot distribution, and I'm wondering > how to best fit the parameters of the distribution given the data. > > I'm using the equation y = P(x + A)^-B as the canonical form of the > Zipf-Mandelbrot curve. > > As a first attempt, I fixed the two endpoints (x=1 and y=1) to coincide with > the expected values on the graph, which analytically gives me values for P > and B as functions of A. Then I twiddle around with values for A until it > looks nice. On the attached graph, this means I end up with A=8. > > This method pretty clearly overfits the endpoints, and is subject to the > whims of my eyeballs for A. What would members of this list suggest for > alternative fitting methods? > > Thanks, > > -Ken > > _______________________________________________ > R-lang mailing list > [email protected] > https://ling.ucsd.edu/mailman/listinfo.cgi/r-lang > -- Marco Baroni CIMeC, University of Trento http://www.form.unitn.it/~baroni _______________________________________________ R-lang mailing list [email protected] https://ling.ucsd.edu/mailman/listinfo.cgi/r-lang
