Plotting goes through qeviz: plot(), edges(), nodes() and units() keep their signatures and return a qe_plot instead of a bundled plotly figure. The 1300-line plotly implementation is gone, and with it the LearnGeom, plotly and scales dependencies; qeviz (>= 0.7.1) is imported. plot(legacy = TRUE) keeps the ONA 0.1 scaling (axes fixed at +/-3, edge widths and node sizes in data units) for figures that must match older ones.
plot() and its helpers work again on sets made since the move to rENA::model(). They assumed the ENA_DIRECTION column of the original ONA 0.1 sets, which rENA::model() does not add; response_rows() / direction_rows() in R/utils.R treat a set without the column as response-only.
model(rotation.set = ) works again. It accepts a fitted ONA set, its $rotation, or any list carrying a rotation.matrix, projects the new accumulation into that space, keeps the fitted node positions, and centres the points unless center_to_origin = FALSE.
Parity suite against the original ONA 0.1.1 package (frozen output in tests/testthat/fixtures/ona_0.1.1.rds, regenerated by data-raw/ona_0.1.1_fixture.R): windowed and HOO accumulations, SVD and means rotations, rotation reuse, line weights, points and nodes agree to 1e-8 with sign tolerance per dimension. A second suite checks model() against the step-by-step rENA pipeline. The case with an empty unit network runs when the installed rENA keeps empty units at the origin (rENA > 0.4.16) and is skipped otherwise.
Requires rENA (>= 0.4.17), whose model() keeps a unit whose network is empty at the origin, left out of the centering mean and the node fit, as ONA 0.1.1 did. The parity suite’s zero-network test (RS.data accumulated with GroupName in the units) runs against it.