Cool! Thanks for sharing.

I would also like to add, Stan is not that hard. Conceptually it's a bit of a jump, especially if you're coming from comp sci not statistics, but I think it's really worth it. Here are some helpful links:

Stan website: http://mc-stan.org/

A primer on Stan: http://www.stat.columbia.edu/~gelman/research/published/stan...

The Python interface PyStan documentation: https://pystan.readthedocs.io/en/latest/

A case using Stan in a real startup's product: https://www.smartly.io/blog/tutorial-how-we-productized-baye...

Another good one: http://dippl.org/

dippl would be a good name for a startup

And it should headed up by a Dipl. Ing. Or a Dipl. Inf.

https://en.wikipedia.org/wiki/Engineer%27s_degree#Germany_an...

A similar package is Edward: https://github.com/blei-lab/edward/blob/master/README.md Which was built on TensorFlow, so it has neural networks goodies, but more importantly is GPU-accelerated! PyMC3 is on the way to using Theano but wasn't done last time I checked.

This reminds me of this discussion about AB testing, power statistics & significance: https://p-value-convergence.herokuapp.com/

Thanks, got me interested in the subject.