A test of fit_shape() used a simulated tree, whose topology differs between arm64 and x86-64 for the same seed; it now reads a fixed tree. ?sim_tree_shape documents that the same seed gives the same trees on the same platform.
treeshapesim 0.2.0
fit_shape() estimates beta (maximum likelihood of the beta-splitting model from the topology) and c (maximum likelihood of the DDX model of laser, with c = -x) from an empirical tree, with an optional parametric bootstrap; on 30 random trees, beta matched the archived apTreeshape (maxlik.betasplit()) and c matched laser (DDX()) to within 3e-5.
sim_tree_target() draws trees whose gamma statistic and normalized Colless index fall within a tolerance of target values: beta and c come from calibration curves (calibrate_shape(), inverted by shape_params()), and topologies and internode intervals are rejected separately, which is equivalent to rejecting whole trees but much more efficient.
shape_grid() crosses levels of gamma and Colless and gives the parameters of each combination.
tree_shape() returns the gamma statistic and the Colless index normalized against its Yule expectation, in pure R (same values as ape::gammaStat() and treestats::colless(normalization = "yule")).
sim_tree_shape() gains nsim (returns a multiPhylo) and root_age; with the defaults it returns the same trees as before.
assign_ages() finds subtree sizes and children in one pass instead of with ape::prop.part(), and checks its arguments; trees are unchanged.
A vignette and a CITATION file.
rtopology_beta() builds the topology iteratively instead of recursively, so large and very imbalanced trees (e.g. 3,000 tips with beta near -2) no longer fail; with the same seed it returns exactly the same trees as before.
sim_tree_shape(), rtopology_beta() and intervals_lineage_rate() check their arguments: n must be a single integer of at least 2, beta a single finite number greater than -2 and c a single finite number (non-finite c used to return trees with NaN branch lengths).
