Algorithmic wish-list

[peter-i-frazier]

It would be very nice to have additional algorithms be implemented in Cornell MOE. This would make it easier for ML researchers doing comparisons against existing methods, and also would be useful for people using Cornell-MOE for real problems. Two algorithms to be added that seem relatively high in priority for solving real problems, and that don't seem to be present in many other packages, are:

• The ability to specify expensive-to-evaluate constraints
• Support for high-dimensional Bayesian optimization, e.g., using random embeddings or other recently published methods

A nice-to-have for ML researchers doing benchmarking would be implementations of entropy search methods.