Back-Calculate Sample Concentrations from Posterior Draws — predict_samples

For each test sample, evaluates the inverse model at every posterior draw to produce a posterior distribution of predicted concentration. The variance definition matches predict_grid_bayes() via include_measurement_error, so the sample pcov is directly comparable to the precision profile.

Usage

predict_samples_bayes(
  samples,
  bayes_fit,
  curve_idx = 1L,
  response_variable,
  is_log_response = TRUE,
  n_draws = NULL,
  cv_x_max = 150,
  pcov_threshold = 20,
  is_log_x = TRUE,
  include_measurement_error = TRUE
)

Arguments

samples: Data frame of test samples.
bayes_fit: Output of fit_bayes_single().
curve_idx: Integer. Which curve (1-based Stan index).
response_variable: Character.
is_log_response: Logical.
n_draws: Integer or NULL.
cv_x_max: Numeric. Default 150.
pcov_threshold: Numeric. Percent CV threshold for pcov_pass. Default 20.
is_log_x: Logical. Default TRUE.
include_measurement_error: Logical. If TRUE (default) inject the SAME observation noise the grid uses before back-calculating, so samples lie on the measurement/CDAN profile. If FALSE, invert the observed response with no added noise (curve/parameter precision only). MUST match the value passed to predict_grid_bayes() for the same fit.

Value

Data frame with original sample columns plus prediction columns.