The geometric feature computation defined in EMPSN is implemented assuming that all cells with rank 0, 1 and 2 have 1, 2, and 3 underlying nodes respectively. Since functional groups, rings and other potential cells of rank 2 may have more (or less) than 3 nodes, we need a new set of geometric features that do not make this assumption.
Furthermore, since there are actually functional groups with 2 nodes, the existing geometric computation breaks (because it looks for the third node for some features like area, which does not exist). To deal with this issue, functional_group_lift presently does not return cells with less than 3 nodes. The PR that implements the new geometric features should thus undo that change as well and remove the cardinality constraint on the outputs of functional_group_lift.
ekarais
commented
7 months ago
The geometric feature computation defined in EMPSN is implemented assuming that all cells with rank 0, 1 and 2 have 1, 2, and 3 underlying nodes respectively. Since functional groups, rings and other potential cells of rank 2 may have more (or less) than 3 nodes, we need a new set of geometric features that do not make this assumption.
Furthermore, since there are actually functional groups with 2 nodes, the existing geometric computation breaks (because it looks for the third node for some features like area, which does not exist). To deal with this issue,
functional_group_liftpresently does not return cells with less than 3 nodes. The PR that implements the new geometric features should thus undo that change as well and remove the cardinality constraint on the outputs offunctional_group_lift.