Thread: Drawing from a semi-log distribution

Hello all. I have a paper that gives the number of binary stars as a function of period. The period is given in a log scale. I want to create at random a list of binary stars, with their period coming from this distribution. The problem is, I am not sure quite how to do this properly with a log distribution. What I currently do is choose at random, based on the cumulative probability, one of the edges of the histogram. I then the actual period is calculated from a uniform random distribution between this edge and the next edge to the left. I always figured this would be an approximation, but it is really scewing the distribution I draw. Can someone please give me some advice? I have attached images of the distribution taken from the paper, a Matlab plot of it (so basically the same thing) and then a histogram of the distribution I draw, which you can see is skewed. Note that the data was taken off the histogram by eye.

Thank you in advance!

Hey aussiedave.

Do you want to sample for an analytic distribution or from an empirical distribution (like the one listed in your graphic)?

Originally Posted by chiro

Hey aussiedave.

Do you want to sample for an analytic distribution or from an empirical distribution (like the one listed in your graphic)?

Hey Chiro,

I want to sample from an empirical distribution. As you can see by comparing the right plot to the other two, I am drawing something that is somewhat similar but just a bit scewed. This is drawn over 10,000 binary stars so the number of statistics is pretty good. Repeated attempts all give me the same approximate result.

Try look at re-sampling technique code in something like R (which is open source):

R Tutorials--Resampling Techniques