boennecd

# computes the log Cholesky decomposition
log_chol <- function (x) {
  L <- chol(x)
  diag(L) <- log(diag(L))
  L[upper.tri(L, TRUE)]
}

# computes the inverse of the log Cholesky decomposition
log_chol_inv <- function(x) {
  n <- round((sqrt(8 * length(x) + 1) - 1)/2)

boennecd

Provides a version of kinship2::kinship that returns a sparse matrix even when there is only one family. A full dense matrix is never stored. The script assigns two functions to the global enviroment when Rcpp::sourceCpp'ed. These are kinship_sparse_R and kinship_sparse which is an only R and mainly C++ implementation, respectilvy. The function can be sourced by calling

download.file("https://gist.githubusercontent.com/boennecd/b51550a8bec2fde6d48f8f3c2d1f69a1/raw",
 cpp_file <- paste0(tempfile(), ".cpp"))

boennecd

Provides a faster version of the kinship2::makefamid function if sourced using Rcpp::sourceCpp. It also provides a new implementation and quite different implementation based largely on C++ called makefamid_new. You can try the functions by calling

download.file("https://gist.githubusercontent.com/boennecd/d42fc0ced67033c222ddc8c95945af00/raw/makefamid-fast.cpp", 
              src_file <- paste0(tempfile(), ".cpp"))
Rcpp::sourceCpp(src_file)

boennecd

# assign function to simulate a number of independent pedigrees along with 
# the observed survival times for the last generation simulated from a 
# mixed generalized survival model.
# 
# Args: 
#   max_depth: maximum depth of the families (number of generations less 
#              one).
#   max_members: roughly maximum number of members in each pedigree in each 
#                family. 
#  sds: standard deviation of the genetic effect. 

boennecd

/// tmb_includes.h
#ifndef TMB_INCLUDES_H
#define TMB_INCLUDES_H

/* TMB */
#include <TMB.hpp>

namespace CppAD {
inline double Value(double const x){
  return x;

boennecd

#include <Rcpp.h>
#include <memory.h>

std::unique_ptr<size_t[]> get_commutation_unequal_vec
  (unsigned const n, unsigned const m, bool const transpose){
  unsigned const nm = n * m, 
            nnm_p1 = n * nm + 1L, 
             nm_pm = nm + m;
  std::unique_ptr<size_t[]> out(new size_t[nm]);
  size_t * const o_begin = out.get();

boennecd

## Essentially identical to 
##    https://github.com/r-hub/rhub-linux-builders/blob/7af50594ab7a3076adb6e52cce5a912d513a9faa/fedora/Dockerfile
## 
## in the start
FROM fedora:30

## Set a default user. Available via runtime flag
RUN useradd docker

RUN dnf install -y \

boennecd

// [[Rcpp::depends(RcppArmadillo)]]
#include <RcppArmadillo.h>

/* d vech(chol(X)) / d vech(X). See mathoverflow.net/a/232129/134083
*
* Args:
*    X: symmetric positive definite matrix.
*    upper: logical for whether vech denotes the upper triangular part.
*/
// [[Rcpp::export]]

boennecd

---
title: Faster Multivariate Normal densities with RcppArmadillo and OpenMP
author: Nino Hardt, Dicko Ahmadou
license: GPL (>= 2)
tags: armadillo openmp featured
summary: Fast implementation of Multivariate Normal density using RcppArmadillo and OpenMP.
---

The Multivariate Normal density function is used frequently
for a number of problems. Especially for MCMC problems, fast 

boennecd

# probit-issue-simulate.R
library(Rcpp)
# 'Weibull_mix.cpp' is at
#   https://gist.github.com/boennecd/770cd7db6f0fa1b6829311ef005436b7
Rcpp::sourceCpp("Weibull_mix.cpp")

# Simulate potentially right censored outcomes from a generalized survival
# model with a given link function and a mixed weibull distribution as the 
# baseline. 
#