bip-pde-vi

Codes for performing variational inference for Bayesian Inverse Problems that involve PDEs.

1. To run MCMC methods, run python mcmc_launcher.py with appropriate flags.
2. To run VB methods, run python vi_launcher.py with appropriate flags.

Note that one needs Python bindings for an FEM solves. The python must be called poisson_1d_fem_driver to be able to perform inference and needs to implement the following methods:

1. set_data_gen_seed(data_seed)
2. set_params(kappa)
3. get_log_lik()
4. get_log_lik_grad()
5. get_param_coords() (this is the coordinates of the centroids of elements for the discretisation of kappa)
6. generate_observations(kappa)
7. get_solution_mean()
8. get_observations()