This is a question that also interests me. I've been thrown into a lot of data sciencey work this year with learning analytics data from MOOCs - so far mostly cleaning and organizing, graphing etc, but we're moving into more machine learning, inference etc. I have a poor math background, but eager to learn, so while I've done most of the practical work in R and now Python with Pandas, I am playing with things like Learn Math, Logic and Computer Science with Haskell (also Think Stats and Think Bayes with Python).

I would love to see a more mature package with IHaskell, easy access to graphing, and a nice Pandas/data.table like library, and a set of statistics tutorials written around them, where you are basically learning statistics/probability at the same time as using the language. (I found this paper on functional probabilistic programming in Haskell very fascinating - the idea of using a Monad to wrap distributions, uncertainty levels etc around numbers seems very powerful http://web.engr.oregonstate.edu/~erwig/papers/PFP_JFP06.pdf)