rollout_reduced

rollout_reduced#

hybrid_aif.inference.rollout_reduced(m, Fs, bs, Q_chols, transition_log_weights, horizon, k)[source]#

Scan horizon steps of propagate -> reduce(k) with static shapes.

transition_log_weights is a static table, (M,) or (K, M), or a per-step schedule, (horizon, M) or (horizon, K, M) (see _as_weight_schedule()); a per-component table needs K == k. The carry is first brought to exactly k components. Returns (final, trajectory), the trajectory stacked on a leading horizon axis.