direct_optimise_replicates: Direct maximum likelihood optimizer for a list of frequency...
In statdivlab/rre: Replicate Richness Estimator

Description Usage Arguments Details Examples

Finds the MLE for a list of replicate frequency count tables drawn from the same population using the full likelihood.

direct_optimise_replicates(fct_list, penalty = NULL, lambda = NULL,
  multiplier = 10, c_seq_len = 100, starts = NULL,
  alpha_min = 1e-10, delta_min = 1e-10, alpha_max = 1e+05,
  delta_max = 1e+05, search_scheme = "grid")

`fct_list`	a list of frequency count tables, see `make_frequency_count_table` for more information on this data structure.
`penalty`	method of penalizing/regularizing the maximum likelihood solution.
`lambda`	parameter for the h1 penalization method.
`multiplier`	Richness values from the observed richness to `multiplier`*(observed richness) will be tested (default: 10).
`c_seq_len`	Number of candidate C values we will maximize over.
`starts`	A dataframe or matrix of optimization starts. See details.
`alpha_min`	box constraints for optimization over `alpha` and `delta`.
`delta_min`	box constraints for optimization over `alpha` and `delta`.
`alpha_max`	box constraints for optimization over `alpha` and `delta`.
`delta_max`	box constraints for optimization over `alpha` and `delta`.
`search_scheme`	method for iterating over candidate C values.

direct_optimise_replicates() searches over candidate C values in a grid spanning from the maximum observed richness (c) to multiplier *c. At each C value we search over alpha, delta using a gradient search (optim with method = "L-BFGS-B"). We use multiple starts (given in the matrix starts). in optimizing as we have no guaruntee of convexity in this problem.