BOPP is a package for automated marginal maximum a posteriori inference (MMAP) based around the probabilistic programming system Anglican. The user only needs to write their model in the same manner as existing Anglican programs and by using the defopt construct instead of defquery, select the variables to be optimized, with the rest marginalized out. It can also be used as a means of exploiting the target source code to improve Bayesian optimization (BO), delivering things such as automatic domain scaling, unbounded optimization, and implicit constraint satisfaction including equality constraints. The key idea is to use a series of code transformations to extract from the original program all the things that are needed to carry out the MMAP problem, such as the target function itself and a program for optimizing the acquisition function subject to the implicit constraints. These are then passed to our other package Deodorant, which uses these to solve the problem probabilistic programs creates for BO. The following paper should be referred to for full algorithmic details and we ask that you cite this paper if you use BOPP in your work.

1. Rainforth, T., Le, T. A., van de Meent, J.-W., Osborne, M. A., & Wood, F. (2016). Bayesian Optimization for Probabilistic Programs. In Advances in Neural Information Processing Systems (NIPS) (pp. 280–288). BIB

   @inproceedings{rainforth-nips-2016,
     title = {{B}ayesian {O}ptimization for {P}robabilistic {P}rograms},
     author = {Rainforth, Tom and Le, Tuan Anh and van de Meent, Jan-Willem and Osborne, Michael A and Wood, Frank},
     booktitle = {Advances in Neural Information Processing Systems (NIPS)},
     year = {2016},
     pages = {280--288}
   }