Renamed options names (especially GP configuration options)
Changed GP configuration value (ntrain_max, gp_train_n_init)
Fixed initial GP initialization (high-posterior-density ascending order)
Removed unused GP configuration code and cleaned up the hyper-param optimization
Changed the configuration of init_N
Removed wrong and duplicate options
Added the new OptimizeResult dictionary to the output of PyBADS
Increased log level for optimization stalling warning to debug
Raise error when a periodic var is passed
opts_N GP parameter optimization, double hp-parameter optimization (opts_N=2) when we detect low GP mean, or high noise, or double-fit configuration. Otherwise, we use single GP optimization,
Fixed the function_logger duplicate point issue. If the noise is heteroskedastic we estimate the noise with the new observations like in PyVBMC otherwise, we store the observation as a new one. When testing the noise or building the final estimate we do not store the data.
ntrain_max,gp_train_n_init)init_Ndebugopts_N=2) when we detect low GP mean, or high noise, or double-fit configuration. Otherwise, we use single GP optimization,function_loggerduplicate point issue. If the noise is heteroskedastic we estimate the noise with the new observations like in PyVBMC otherwise, we store the observation as a new one. When testing the noise or building the final estimate we do not store the data.bads.optimize()