Reviews: Bayesian Optimization for Probabilistic Programs

It seems that the paper needs a bit of work to improve on this. The results comparing to other Bayesian optimization methods (figure 5) are quite impressive but more useful insights should be provided as to why this is the case. For example when performing hyper-parameter optimization in a graphical model while marginalizing all the other latent variables, or using a structured prediction problem. This will provide the reader with a better understanding of how useful the framework is. However, the introduction of such techniques into PPS give rise to several difficulties, namely (a) dealing with problem-independent priors, (b) unbounded optimization and (c) implicit constraints. The paper seems to address these in an effective way.