In principle, approxposterior should be able to perform Bayesian optimization, given the GP model, to find maximum likelihood solutions. This is obviously not a new idea and can be implemented using george, e.g. this example, but it would be a good tool to joint the approxposterior ecosystem. For example, one could wish to plot the maximum likelihood estimation on a corner plot of approxposterior constraints, or maybe someone is only interested in the "best fit" example. Either way, it could be a nice addition.
A great algorithm to implement would be Jones+1998.
In principle, approxposterior should be able to perform Bayesian optimization, given the GP model, to find maximum likelihood solutions. This is obviously not a new idea and can be implemented using george, e.g. this example, but it would be a good tool to joint the approxposterior ecosystem. For example, one could wish to plot the maximum likelihood estimation on a corner plot of approxposterior constraints, or maybe someone is only interested in the "best fit" example. Either way, it could be a nice addition.
A great algorithm to implement would be Jones+1998.