match_with_spiked_wishart: Compute what spiked SD values will give you the desired top...

In dependentsimr: Simulate Omics-Scale Data with Dependency

Compute what spiked SD values will give you the desired top eigenvalues by iteratively solving

Description

Compute what spiked SD values will give you the desired top eigenvalues by iteratively solving

Usage

match_with_spiked_wishart(
  desired_eigenvalues,
  rank,
  num_observations,
  num_variables,
  population_sd = 1,
  num_iterations = 20,
  num_samples_per_iter = 300
)

Arguments

desired_eigenvalues	Vector of top eigenvalues of sample covariance matrix that should be approximated
rank	number of 'spiked' dimensions to fit
num_observations	Number of observations (samples, columns) in the data matrix
num_variables	Number of variables (features, rows) in the data matrix
population_sd	Standard deviation for allnon-spiked dimensions
num_iterations	Number of iterations to perform to to optimize. Increase to get a closer fit
num_samples_per_iter	Number of eigenvalues to samples to estimate mean eigenvalues for each iteration. Increase to get a closer fit

Value

Fit values of spiked SDs that approximately match the desired eigenvalues

Examples

# First, simulate a dataset with known true values to see if we can get close
true_spiked_sd <- c(500, 100)
num_variables <- 1000
num_observations <- 10-1
sampled_eigenvalues <- sample_spiked_wishart(true_spiked_sd, num_observations,
      num_variables, num_eigs=2)
# Fit some spiked SDs that give eigenvalues similar to sampled_eigenvalues
fit <- match_with_spiked_wishart(
    sampled_eigenvalues,
    rank=2,
    num_observations,
    num_variables,
    num_iterations=5,
    num_samples_per_iter=50
)
# fit$spiked_sd should now be close to giving the specified sampled_eigenvalues (in expectation)
# fit$spiked_sd won't match the original true_spiked_sd too closely since it only
# fits the single sample # that gave  sampled_eigenvalue
# fit$population_sd gives fit value for the population SD

dependentsimr documentation built on Aug. 8, 2025, 6:23 p.m.