This is a long time project of mine (started in 2018) where I had an idea that it is possible to do exact inference of probabilistic programs even when loops (including infinite loops) are involved. I didn't publicize it too much as performance was too abysmal to be practical, but recently with encouragement from a friend I decided to add more features (support integers!) and polish it up.

A probabilistic programming language isn't really a general purpose language; instead it uses a programming language to describe a random process or phenomenon, and the inference tool reports a distribution of outcomes.

Here is an example from the README that can be solved with my tool. Suppose there are four cars and eight spaces. Initially the cars are located in the left half of the spaces. At each iteration, each car flips a fair coin. When the coin is heads and if there is a space in front, then the car moves forward by one space. How many iterations are needed to get all four cars in the right half? It turns out that the most likely outcome is 17 iterations, and the minimum number of iterations, seven, has a probability of only 1/65536.

I don't have a whole lot of use cases, but I found it really fun to hack on this project!