Grid so that 2*ln(2)*quantile(d_ij)/factor_up_alpha < t_1/2 < factor_down_alpha * ln(2) * h_tree, with t_1/2 the phylogenetic half life: t_1/2 = log(2)/alpha. Ensures that for alpha_min, it is almost a BM, and for alpha_max, almost all the tips are decorrelated.
Usage
find_grid_alpha(
phy,
alpha = NULL,
nbr_alpha = 10,
factor_up_alpha = 2,
factor_down_alpha = 3,
quantile_low_distance = 1e-04,
log_transform = TRUE,
allow_negative = FALSE,
...
)Arguments
- phy
phylogenetic tree of class "
phylo"- alpha
fixed vector of alpha values if already known. Default to NULL.
- nbr_alpha
the number of elements in the grid
- factor_up_alpha
factor for up scalability
- factor_down_alpha
factor for down scalability
- quantile_low_distance
quantile for min distance
- log_transform
whether to take a log scale for the spacing of alpha values. Default to TRUE.
- allow_negative
whether to allow negative values for alpha (Early Burst). See documentation of
PhyloEMfor more details. Default to FALSE.- ...
not used.
Details
If quantile_low_distance=0, then quantile(d_ij)=min(d_ij), and, for any
two tips i,j, the correlation between i and j is bounded by exp(-factor_up_alpha/2).
Those values of alpha will be used for the re-scaling of the tree, which has an
exponential term in exp(2*alpha*h). The function makes sure that this number is
below the maximal float allowed (equals to .Machine$double.xmax).