matflow.subset_simulation.subset_simulation.generate_next_level_samples_DA#
- matflow.subset_simulation.subset_simulation.generate_next_level_samples_DA(performance, performance_coarse, num_chains, num_states, dimension, chain_seeds, chain_g, level_idx, master_seed, threshold, proposal, log_surrogate_weight, transformation=None, temperature=1.0, num_inner_states=1, spawn_key=None, debug=False)#
Generate one conditional level using fixed-subchain RST.
This is the randomised-length subchain surrogate transition (RST) algorithm but with a fixed length subchain. For num_inner_states=1, this should be identical to
generate_next_level_samples_DA_single_inner.Each outer transition:
- current x
v
- num_inner_states coarse-MH steps
v
- endpoint psi
v
- one fine evaluation (unless psi == x)
v
- fine correction
v
new x
The inner MH kernel targets the coarse surrogate
pi_C(x) ∝ phi(x) * log_surrogate_weight(g_c(x)).
The endpoint is then corrected to the fine target.