boot_specieslevel_n: Bootstrap metric at the species level multiple times.

In valentinitnelav/bootstrapnet: Bootstrap Network Metrics

Bootstrap metric at the species level multiple times.

Description

Bootstrap a single network of interactions n times in parallel. This function is used for network metrics at the species level. Starts with a small sample size of interactions (e.g. start = 30), builds the corresponding web matrix/network, computes its metric (e.g. index = "d") using specieslevel, then adds new interactions (e.g. step = 20) until all interactions are sampled. The last sample is actually the entire network. Repeats n times (as given in n_boot) these steps in parallel on multiple CPUs.

Usage

boot_specieslevel_n(
  data,
  col_lower,
  col_higher,
  index,
  level,
  start,
  step,
  n_boot,
  n_cpu,
  ...
)

Arguments

data	Data frame of interactions from which to build and sample web matrices. Each interaction (row in the data frame) must be repeated as many times as it was observed. E.g. if species_1 x species_2 was observed 5 times, then repeat that row 5 times within the data frame. See examples below.
col_lower	Quoted column name in data for lower trophic level species (plants).
col_higher	Quoted column name in data for higher trophic level species (insects).
index	Passed to specieslevel. See ?bipartite::specieslevel for details.
level	Passed to specieslevel. See ?bipartite::specieslevel for details. For which level should the level-specific indices be computed: 'both' (default), 'lower' or 'higher'?
start	Integer. The sample size (number of interactions) to start the bootstrap with. If the start sample size is small (e.g. 5 or 10), then first iterations might results in NaN-s and warning messages are displayed. Consider to set start to maybe 10\% of your total unique interactions.
step	Integer. Sample size (number of interactions) used to increase gradually the sampled network until all interactions are sampled. If step is too small (e.g. 1) then the computation time is very long depending on your total number of interactions from which samples are taken. Consider to set step to maybe 5-10\% of your total unique interactions.
n_boot	Number of desired bootstraps (50 or 100 can be enough).
n_cpu	Number of CPU-s to use for parallel processing.
...	Other arguments passed to specieslevel like logbase, etc.

Value

Returns a list of 1, 2 or 4 arrays of matrices. The species names are stored as the row names of each matrix. The number of columns of a matrix indicates how many iterations took place. This is decided internally based on the given values to start, step and the total number of rows (interactions) in data. The column names give the sample size at each iteration. The last iteration (last column name) is always the entire network (total number of interactions in data). The 3rd dimension (number of matrices in the array) corresponds to n_boot (number of bootstraps).

Examples

library(bootstrapnet)
library(magrittr)
library(bipartite)
data(Safariland)

Safariland %>%
  web_matrix_to_df() %>%
  boot_specieslevel_n(col_lower = "lower", # column name for plants
                      col_higher = "higher", # column name for insects
                      index = "d",
                      level = "both",
                      start = 100,
                      step = 100,
                      n_boot = 10,
                      n_cpu = 2)

valentinitnelav/bootstrapnet documentation built on June 5, 2024, 3:21 p.m.