staple_bin_mat: STAPLE on binary matrix

In neuroconductor/stapler: Simultaneous Truth and Performance Level Estimation

Description

STAPLE on binary matrix

Usage

staple_bin_mat(
  x,
  sens_init = 0.99999,
  spec_init = 0.99999,
  max_iter = 10000,
  tol = .Machine$double.eps,
  prior = "mean",
  verbose = TRUE,
  trace = 10,
  drop_all_same = FALSE
)

Arguments

x	a nxr matrix where there are n raters and r elements rated
sens_init	Initialize parameter for sensitivity (p)
spec_init	Initialize parameter for specificity (q)
max_iter	Maximum number of iterations to run
tol	Tolerance for convergence
prior	Either "mean" or a vector of prior probabilities,
verbose	print diagnostic messages
trace	Number for modulus to print out verbose iterations
drop_all_same	drop all records where they are all the same. DO NOT use in practice, only for validation of past results

Value

List of output sensitivities, specificities, and vector of probabilities

Examples

n = 5
r = 1000
sens = c(0.8, 0.9, 0.8, 0.5, 0.8)
spec = c(0.9, 0.75, 0.99, 0.98, 0.92)
suppressWarnings(RNGversion("3.5.0"))
set.seed(20171120)
n_1 = 200
n_0 = r - n_1
truth = c(rep(0, n_0), rep(1, n_1))
pred_1 = rbinom(n = n, size = n_1, prob = sens)
pred_0 = rbinom(n = n, size = n_0, prob = spec)
pred_0 = sapply(pred_0, function(n) {
   sample(c(rep(0, n), rep(1, n_0 -n)))
})
pred_1 = sapply(pred_1, function(n) {
   sample(c(rep(1, n), rep(0, n_1 -n)))
})
pred = rbind(pred_0, pred_1)
true_sens = colMeans(pred[ truth == 1, ])
true_spec = colMeans(1-pred[ truth == 0, ])
x = t(pred)
staple_out = staple_bin_mat(x)
testthat::expect_equal(staple_out$sensitivity,
c(0.781593858553476, 0.895868301462594,
0.760514086161722, 0.464483444340873,
0.765239314719065))
staple_out_prior = staple_bin_mat(x, prior = rep(0.5, r))
testthat::expect_equal(staple_out_prior$sensitivity,
c(0.683572080864211, 0.821556768891859,
0.619166852992802, 0.389409921992467, 0.67042085955546))

neuroconductor/stapler documentation built on May 20, 2021, 6:39 p.m.