trustOptim is an R package for unconstrained optimization of nonlinear objective functions with sparse Hessians. You should consider using this package if:
A common use case for a sparse Hessian is a hierarchical model that assumes conditional independences across heterogeneous units. Even if the Hessian is not sparse, and/or it is hard to compute the Hessian analytically, this package does support BFGS and SR1 updates instead. But the real benefit of this package comes from exploiting the sparsity of the Hessian of the objective function.
The latest release version is available on CRAN. You should be able to install it with a simple
install.packages("trustOptim")`.
The source code is available at https://github.com/braunm/trustOptim .
The trust.optim
function calls the optimizer.
The objective function is defined by three R functions, all of which take a numeric vector as the first argument.
Starting from parameter vector x, to minimize a function f, call
opt <- trust.optim(x, f, grad, hess, method="Sparse")
See the vignettes for examples, and the package manual for details, options, etc.