Solver objective functions

Function factory producing functions for use in optimization. These are objective function and its gradient. They can be used, e.g., as parameters fn and gr of stats::optim(). Note that the functions compute their values through objective(), however, calls to the latter are memoised, so actual solver should never be called twice for the same data (x). In particular, computing value and gradient for a given x requires only a single solver run. To obtain non-memoised functions use make_functions() composed with objective() (see example in make_functions()).

objective_functions(solver, data, ...)

Arguments

solver	object of class `solver`.
data	observed ('exact') data.
...	additional args passed to `objective()` or `memoise::memoise()`, note that some solvers can have some mandatory parameters (see `required_args()`), e.g. solvers created with `adaptive_shell_solver()` require `precision`.

Value

List with one or two components:

value: objective value function,
gradient: objective gradient function, missing if solver does not compute Jacobian matrix.

Examples

s <- fake_adaptive_solver(4, 5)
result <- run(s, c(10, 10, 10, 10), precision = 5.0, silent = TRUE)
observed_data <- result$qoi
x <- c(10.5, 9.44, 10.21, 8.14)
solver_funs <- objective_functions(s, observed_data, precision = 30.0,
silent = TRUE)
solver_funs$value(x)
#> [1] 2594156034
solver_funs$gradient(x)
#> [1] -6208209534 -4059190749 -5551351184 -2246937119

Arguments

Value

See also

Examples