propagate

Contents

propagate#

hybrid_aif.inference.propagate(m, Fs, bs, Q_chols=None, transition_log_weights=None)[source]#

Exact grow: K components through M transitions -> K*M components.

Fs (M, dim, dim), bs (M, dim), Q_chols (M, dim, r), any factors of the process noise; transition_log_weights (M,) shared or (K, M) per-component, added to the component log-weights as given. bs may instead be a ConditionalGaussian with parent_shape (M,) supplying offsets and noise factors, with Q_chols then None. Output is component-major: slot k*M + j is component k through mode j. Batch axes of m broadcast through.