In Florian-40/mypkgr: My 1st package

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(mypkgr)
library(devtools)
devtools::install_github("Florian-40/microbenchmark")
library(microbenchmark)
library(ggplot2)

Comparison with our microbenchmark package.

You can find microbenchmark package on my Git-Hub page : https://github.com/Florian-40/microbenchmark.git

mb <- microbenchmark(mvtnorm::dmvnorm(rep(1.96, 2)),
                     mvnpdf(x=matrix(rep(1.96,2)), Log=FALSE),
                     times=1000L)

mb
expression<-c('dmvnorm', 'mvnpdf')
autoplot(mb, expression)

And with multidimensionnal, we get :

n <- 100
mb <- microbenchmark(mvtnorm::dmvnorm(matrix(1.96, nrow = n, ncol = 2)),
                     mvnpdf(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     times=100L)
mb
expression<-c('dmvnorm', 'mvnpdf')
autoplot(mb, expression)

Profiling

By using this lines in terminal, we can check profiling of the code :

profvis::profvis({
+ n <- 10e4
+ pdfval <- mvnpdf(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE)
+ })

mvnpdfsmartfunction.

We can use new function mvnpdfmart and check profile with similar code :

profvis::profvis({
+ n <- 10e4
+ pdfval <- mvnpdfsmart(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE)
+ })

We can compare results with a microbenchmark :

n <- 1000
mb <- microbenchmark(mvtnorm::dmvnorm(matrix(1.96, nrow = n, ncol = 2)),
                     mvnpdf(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdfsmart(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     times=100L)
mb
expression<-c('dmvnorm', 'mvnpdf', 'mvnpdfsmart')
autoplot(mb, expression)

mvnpdfoptim function.

We can use new function mvnpdfmart and check profile with similar code :

profvis::profvis({
+ n <- 10e4
+ pdfval <- mvnpdfoptim(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE)
+ })

We can compare all results with a microbenchmark:

n <- 1000
mb <- microbenchmark(mvtnorm::dmvnorm(matrix(1.96, nrow = n, ncol = 2)),
                     mvnpdf(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdfsmart(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdfoptim(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     times=100L)
mb
expression<-c('dmvnorm', 'mvnpdf', 'mvnpdfsmart', 'mvnpdfoptim')
autoplot(mb, expression)

Using Rccp Armadillo functions.

We write a better inversion matrix with C++ language and we include it in mvnpdfsmart. Thus, we get microbenchmark :

n <- 1000
mb <- microbenchmark(mvtnorm::dmvnorm(matrix(1.96, nrow = n, ncol = 2)),
                     mvnpdf(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdfsmart(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdfoptim(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdf_invC(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     times=100L)
mb
expression<-c('dmvnorm', 'mvnpdf', 'mvnpdfsmart', 'mvnpdfoptim', 'mvnpdf_invC')
autoplot(mb, expression)

Now, use complete implementation of mvnpdfsmart in C++.

Rcpp::sourceCpp("../src/mvnpdfC.cpp")
n <- 1000
mb <- microbenchmark(mvtnorm::dmvnorm(matrix(1.96, nrow = n, ncol = 2)),
                     mvnpdf(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdfsmart(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdfoptim(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdf_invC(x=matrix(1.96, nrow = 2, ncol = n), Log=FALSE),
                     mvnpdfC(x=matrix(1.96, nrow = 2, ncol = n), mean = rep(0, 2), varcovM = diag(2), Log=FALSE),
                     times=100L)
mb
expression<-c('dmvnorm', 'mvnpdf', 'mvnpdfsmart', 'mvnpdfoptim', 'mvnpdf_invC', 'mvnpdfC')
autoplot(mb, expression)

Parallel code

1st simple example.

Let's try to calculate the logarithm of $n$ numbers.

mb <- microbenchmark(log_par(1:100), log_seq(1:100), times=50)
mb
expression <- c('log_par', 'log_seq')
autoplot(mb, expression)

Florian-40/mypkgr documentation built on April 30, 2022, 6:10 p.m.