Currently in SMC, when data arrive as pairwise preferences, all particles get the same initial value. In contrast, with partial ranks, they are randomly assigned.
Relevant code lines are here. With pairwise preference data, missing_indicator is always empty and the data generate for setup_rank_data are used for each particle.
Currently in SMC, when data arrive as pairwise preferences, all particles get the same initial value. In contrast, with partial ranks, they are randomly assigned.
Relevant code lines are here. With pairwise preference data,
missing_indicatoris always empty and the data generate forsetup_rank_dataare used for each particle.https://github.com/ocbe-uio/BayesMallows/blob/6f6b85d81b2c5249d155a12f35306810722359a9/src/particles.cpp#L16-L49