cv_replicates: Method 2 and Method 4
In statdivlab/rre: Replicate Richness Estimator

Description Usage Arguments Details Examples

Method 2 is cross-validation using the likelihood in the evaluation step. Method 4 is cross-validation using the goodness of fit statistic in the evaluation step.

cv_replicates(fct_list, lambda_vec = seq(0, 20, by = 2),
  starts = data.frame(alpha = c(0.01, 0.01), delta = c(0.01, 1e-04)),
  partitions = 10, eval_function = "gof_chi_sq", multiplier = 20,
  c_seq_length = 96, ...)

`fct_list`	A list of frequency count tables, assumed to be replicates.
`lambda_vec`	The values of the penalty parameter we consider in selecting λ.
`starts`	Starting values for `alpha` and `delta` in the MLE procedure.
`partitions`	An integer indicating the number of times to randomly split the data into testing and validating subsets.
`eval_function`	A function which evaluates how well a set of parameters fit a list of frequency count tables. To conform to goodness of fit, we use the negative of the likelihood function so that low scores are better.
`multiplier`	The upper bound of the grid of candidate C values, stated in terms of a multiple of the maximum observed richess (c). For example if c is 50 and multiplier is 10, the method evaluates the likelihood in a C grid from 50 to 500.
`c_seq_len`	The number of points in the C grid search.

Methods 2 and 4 have very similar structure we we've included them both in the same function. To run each method use:

Method 2: cv_replicates(..., "neg_unreg_like")
Method 4: cv_replicates(..., "gof_chi_sq")

In each method we partition the data partitions times into training and evaluation subsets. An estimate for each λ in lambda_vec is generated and we evaluate them using the evaluation subset. The evaluation step depends on the method, see paper or source code for details of how these functions work.