Struggling to generate initial values when probabilities are low

[dbarneche]

This will fail

library(bayesnec)
library(tidyverse)
b_top <- 100
b_beta <- -0.2
b_nec <- 4
fct <- function(...)as.integer(round(bayesnec:::pred_nec3param(...)))
dat <- data.frame(x = seq(1, 20, length.out = 10), tr = 500) %>%
  dplyr::mutate(y = fct(b_beta, b_nec, b_top, x))
test <- bnec(y | trials(tr) ~ model(x, "nec3param"), data = dat, family = binomial(link = "identity"))

But this will work just fine

dat$tr <- 100
test <- bnec(y | trials(tr) ~ model(x, "nec3param"), data = dat, family = binomial(link = "identity"))

So I suspect that the current blind sampling of make_inits is not good enough when probabilities are low.