design/TMB/check_autodiff.R

In openpharma/mmrm: Mixed Models for Repeated Measures

## Check correctness of 'autdiff' namespace
library(TMB)

## Compile and load the model
compile("check_autodiff.cpp")
dyn.load(dynlib("check_autodiff"))

## Data and parameters
data <- list()
parameters <- list(theta = 1:2)

## Make a function object
data$select <- 1
obj <- MakeADFun(data, parameters, DLL = "check_autodiff", ADreport = TRUE, silent = TRUE)

obj$report(obj$par) # gives g and h
obj$gr(obj$par)
obj$he(obj$par)
test <- sdreport(obj)
str(test) # gives f

## Check autodiff::gradient and autodiff::hessian
stopifnot(all(obj$report(obj$par)$g == obj$gr(obj$par)))
stopifnot(identical(obj$report(obj$par)$h, obj$he(obj$par)))

## Make a function object
data$select <- 2
obj <- MakeADFun(data, parameters, DLL = "check_autodiff", ADreport = TRUE, silent = TRUE)
stopifnot(all(obj$gr(obj$par) == diag(exp(obj$par))))

## Make new function object
data$select <- 3
obj <- MakeADFun(data, parameters, DLL = "check_autodiff", ADreport = TRUE, silent = TRUE)
stopifnot(all(obj$report(obj$par)$j == obj$gr(obj$par)))