I just noticed that the fit_smc! API doesn't actually have an ESS chooser, like doing the fitting purely incrementally (including maybe_resample!). If resampling is switched on then its fixed at ess $ =N_p /2 $
|
ess_threshold = adaptive_resampling ? n_particles/2 : n_particles |
It would be worth updating this? Maybe have as just ess as a numeric and if thats negative then thats equivalent to switching resampling off?
I just noticed that the
fit_smc!API doesn't actually have an ESS chooser, like doing the fitting purely incrementally (includingmaybe_resample!). If resampling is switched on then its fixed at ess $ =N_p /2 $AutoGP.jl/src/inference_smc_anneal_data.jl
Line 231 in 4665511
It would be worth updating this? Maybe have as just ess as a numeric and if thats negative then thats equivalent to switching resampling off?