fpi#
- hybrid_aif.inference.fpi(log_likelihoods, A_deps, priors, observations, num_iters=16, masks=None)[source]#
Mean-field posteriors over state factors given one observation per modality.
fpi_support()is the same sweep, also returning the per-factorunsupportedflags. Every log term is clipped atLOG_ZEROon the way in, so-infentries are legal and an observation no state can explain gives a finite posterior rather thanNaN. The returned logits are unnormalised (messages plus log prior).masks, one entry per modality, marks which observations are present; a masked-out modality contributes the zero message. A PythonFalseskips the modality statically and never reads its observation (which may beNone); a traced flag selects at run time, so padding of the right shape must be there.Design: docs/design/zero-support-evidence.md Z4