TODO.md

In mlysy/LocalCop: Local Likelihood Inference for Conditional Copula Models

29/01/2024

• [ ] Finish documentation for d/p/h of {gumbel/frank/clayton}.hpp follwing template.

• [ ] Put all of these directly on the log scale, i.e., if(giv_log) return ans else return exp(ans).

• [ ] Write d/p/h for Gaussian copula in the style of the previous copulas. This also involves checking that R and C++ version agree. See test/testthat for how it was done on previous copulas.

• [ ] Write pt() and qt() in preparation for student t copula. Note: this will be done by Martin.

• [ ] Write d/p/h for Student-t copula in style of the others.

27/04/2021

Compile TMB code

After adding new TMB files, compile as follows:

# setwd to within your source directory
TMBtools::export_models() # tell package to recognize TMB code has changed.
pkgbuild::compile_dll() # compile package TMB/C++ code.

# reinstall package
devtools::document() # sets up documentation and namespace
devtools::install()

Tip: Always quit + restart R at this juncture. TMB C++ code is compiled into a separate so from the rest of the package i.e., that which Rcpp code lives in. This non-standard compiling mechanism is not guaranteed to work without a fresh R session.

R API

require(LocalCop)

clayton_nll <- TMB::MakeADFun(
  # fixed inputs
  data = list(
    model = "ClaytonNLL",
    u1 = u1, 
    u2 = u2,
    weights = weights,
  ),
  # variable inputs
  parameters = list(
    # initial value to optimizer
    theta = theta0
  ),
  # config
  silent = TRUE, # for debugging
  DLL = "LocalCop_TMBExports"
)

# ClaytonNLL function and its derivatives
clayton_nll$fn(theta) # negative of weighted loglikelihood;
clayton_nll$gr(theta) # its gradient,
clayton_nll$he(theta) # hessian.

Testing

The tests/testthat folder contains a test that can be modified to test other copula distributions.

mlysy/LocalCop documentation built on Sept. 29, 2024, 9:13 a.m.