I should add utility functions to help users initialize their training sets. That is, given hard bounds for parameter ranges, functions that generate sets of initial conditions for the user's foward model to build the initial GP training set. Options will include sampling from the prior (already implemented), uniform over the ranges, and Latin hypercube sampling as was used in this paper and this paper by Simeon Bird and collaborators. The latin hypercube sampling option can be particularly useful, it seems.
After implementing these utility functions, I should add tests and documentation/examples.
Implemented in the dev branch using the pyDOE latin hypercube sampling implementation, but with the user able to scale by the parameter bounds of their choosing.
I should add utility functions to help users initialize their training sets. That is, given hard bounds for parameter ranges, functions that generate sets of initial conditions for the user's foward model to build the initial GP training set. Options will include sampling from the prior (already implemented), uniform over the ranges, and Latin hypercube sampling as was used in this paper and this paper by Simeon Bird and collaborators. The latin hypercube sampling option can be particularly useful, it seems.
After implementing these utility functions, I should add tests and documentation/examples.