Updated version as accepted at Journal of Statistical Software
Explicit registration of native routines, as required by R 3.4.0.
Added implementation of the complex step method.
Revised vignette.
The ‘direct’ argument to the sparseHessianFD initializer was removed (defunct).
An even more major rewrite of the package. All ACM code was removed, and replaced with original R/C++ implementations.
The sparseHessianFD class is now implemented as an R reference class, and not as an Rcpp module. The sparseHessianFD.new function is deprecated. Instead, use sparseHessianFD to initialize an object. Initialization once again takes place in a single step.
The ‘direct’ computation method has been removed. All computation uses the ‘indirect’ triangular substitution method. The ‘direct’ argument in the initializer for the sparseHessianFD class is now deprecated, and remains solely for compatibility with older versions of the package.
There is a new vignette with a lot more detail about what the package does and how it works.
New matrix helper functions Matrix.to.Pointers and Coord.to.Pointers. The Coord.to.Pattern.Matrix function is deprecated. Use the spMatrix or sparseMatrix functions in the Matrix package instead.
With the removal of ACM-copyrighted code, this package is now licensed under the MPL 2.0.
Essentially a complete re-write of the package.
New vignette, using a binary choice model as an example. Functions for this model are in binary.R. Access sample simluated data with data(binary).
Documentation written using roxygen2
Added unit tests using testthat
Core class has been renamed sparseHessianFD. Construction and initialization are now two separate steps.
New function sparseHessianFD.new is a wrapper around the construction and initialization steps.
new.sparse.hessian.object. Use sparseHessianFD.new instead.Coord.to.Sym.Pattern.Matrix. Use Coord.to.Pattern.Matrix instead, with option symmetric=TRUE.Sym.CSC.to.Matrix. Use the spMatrix function in the Matrix package instead.