Thank you for a providing this great library. The convenience of Python and the speed of native code really is a happy marriage.
I'm missing one thing though. It would be great if you could bias the random jump vector for the PageRank algorithm. This modification is called "personalization". By providing non-uniform jump probabilities you selectively downplay or raise the importance of certain nodes. In natural language processing, where I work, this trick has quite a few applications.
For reference, the parameterization is implemented in NetworkX . Unfortunately, it becomes very slow once the graphs get large.