Simulate Effect of Sample Size on a Projection Pursuit Index
Source:R/sample_size_effect.R
ppi_samplesize_effect.RdFor a given index function, simulates how the index behaves across a range of sample sizes. By default, standard bivariate Gaussian noise is used, but the user may instead supply a 2-column dataset to sample from.
Usage
ppi_samplesize_effect(
index_fun,
sample_sizes = 100,
n_sim = 100,
data = NULL,
csv_file = NULL,
index_args = list(),
workers = 1
)Arguments
- index_fun
A function returning a numeric scalar index. It may accept either a 2-column matrix/data frame, or two numeric vectors (
x,y).- sample_sizes
Numeric vector (e.g., 100, c(50,100,200), seq(30,200,5)). Default is 100.
- n_sim
Integer. Number of simulations per sample size.
- data
Optional 2-column data frame, tibble, or matrix to sample from. If
NULL, standard bivariate Gaussian noise is used.- csv_file
Optional file path to save results as CSV.
- index_args
A named list of additional arguments passed to
index_fun.- workers
Integer. Number of parallel workers to use. Default is 1.
Value
A tibble with:
sample_size: sample size used for each simulation blockpercentile95: 95th percentile of the index values over simulations
Examples
if (FALSE) { # \dontrun{
ppi_samplesize_effect(
cassowaryr::sc_stringy,
sample_sizes = c(100, 300),
n_sim = 3,
index_args = list(binner = "hex"),
workers = 2
)
ppi_samplesize_effect(
cassowaryr::sc_skinny,
sample_sizes = c(100, 300),
n_sim = 3,
data = data_gen("polynomial", degree = 2),
workers = 2
)
} # }