[Numpy-discussion] Vectorize leaks

[Chris Fonnesbeck]

I have narrowed a memory leak in PyMC down to the vectorize() function 
in numpy. I have a simple inverse logit transformation function:

invlogit = lambda x: 1.0 / (1.0 + exp(-1.0 * x))

which runs without leaking when used iteratively during simulations. 
However, when I try to vectorize it, the process' rsize grows each 
iteration of the simulation.

Using a recent (<2 days old) svn build of numpy on OS X 10.4.

C.