Note

Please open your R-Studio as an administrator.

Overview

In this tutorial, we introduce a package to use multiple cores of your system.

After this tutorial you will be able to:

Train a neural network model without using multicores
Train a neural network model without using multicores
Compare results

Using R Markdown

The tutorial that follows was created using R Markdown. As an exercise in using R Markdown, here, we ask you to save your work in an R Markdown file. By the end of this session, you will have recreated the .Rmd file of this tutorial. You don’t have to retype the instruction and comments. Focus only on the R code and add notes that are useful for your own understanding.

To create an R Markdown file:

Choose File -> New File -> R Markdown… from RStudio menu.
Save this file with the name Rcode.Rmd in your working directory. Don’t change the other R Markdown options.
From R studio menu use Code -> Insert Chunk to add boxes to save R code.

You may use the R Markdown cheat sheet https://www.rstudio.com/wp-content/uploads/2015/02/rmarkdown-cheatsheet.pdf to further format the output of the R Markdown file.

R Markdown setup

This is generated by default when we create an R Markdown file.

# Change this to match the directory you have stored your scripts and data
root_directory <- "/media/sanju/B47A9D517A9D10EA/sanju-personal/rcode-clinic/rcode"

knitr::opts_chunk$set(echo = TRUE)
# knitr::opts_knit$set(root.dir = root_directory)

Required packages

Uncomment and install a package if you didn’t already.

#install.packages("caret")
library(caret)

## Loading required package: lattice

## Loading required package: ggplot2

install.packages("doMC", repos="http://R-Forge.R-project.org")

## Installing package into 'C:/Users/Justin Millar/Documents/R/win-library/3.6'
## (as 'lib' is unspecified)

## installing the source package 'doMC'

require(doMC)

## Loading required package: doMC

## Loading required package: foreach

## Loading required package: iterators

## Loading required package: parallel

## For Windows
# install.packages("doParallel")
require(doParallel)

## Loading required package: doParallel

#install.packages("e1071")
#library(e1097)

Downloading Data

The data we are going to use a csv file (having gene expressions for patients) and clinical data (having response) from a breast cancer study. The data are available in folder “session-1” of this link (https://github.com/sanjaysinghrathi/Rcode-clinic/blob/master/Data.csv).

The data file having .csv extension includes the gene expression data for 19 samples (patients) as well as includes clinical parameters like response referring to the samples.

print("Welcome to Session 1")

## [1] "Welcome to Session 1"

Load data

Use this code to read the data.csv file into R directly from Github. Note that will not save the data to your computer. Alternatively, you can download the file using the link above and read it from your local directory.

## Load data file
data <- read.csv("https://raw.githubusercontent.com/sanjaysinghrathi/Rcode-clinic/master/Data.csv", header = TRUE, row.names = 1)

Train the neural network model

Put the timer on to check the time used by model for training.

start_time <- Sys.time()
numFolds <- trainControl(method = 'LOOCV', allowParallel = TRUE, verbose=TRUE , search = "grid")
grid <- expand.grid(size=c(seq(from = 1, to = 10, by = 1)),
                    decay=c(seq(from = 0.0, to = 0.9, by = 0.1)))
set.seed(567)
garbage <- capture.output(suppressWarnings(model <- train(response ~ ., data, method='nnet', trace = FALSE, preProcess = c('center', 'scale'), metric="Accuracy", trControl = numFolds, linout=FALSE, tuneGrid=grid)))
max(model$results$Accuracy, na.rm = TRUE)

## [1] 1

training_time <- Sys.time()
print("Training time...")

## [1] "Training time..."

training_time-start_time

## Time difference of 1.620343 mins

Question: How much time is used by the model to train?

Set the number of cores for DoMC

Find the number of cores available on your computer.

registerDoMC(cores=6)

## For Window users
#registerdoParallel(cores=6)

Train the neural network model with maximum possible cores on computer

Put the timer on to check the time used by model for training.

start_time <- Sys.time()
numFolds <- trainControl(method = 'LOOCV', allowParallel = TRUE, verbose=TRUE , search = "grid")
grid <- expand.grid(size=c(seq(from = 1, to = 10, by = 1)),
                    decay=c(seq(from = 0.0, to = 0.9, by = 0.1)))
set.seed(567)
garbage <- capture.output(suppressWarnings(model <- train(response ~ ., data, method='nnet', trace = FALSE, preProcess = c('center', 'scale'), metric="Accuracy", trControl = numFolds, linout=FALSE, tuneGrid=grid)))
max(model$results$Accuracy, na.rm = TRUE)

## [1] 1

training_time <- Sys.time()
print("Training time...")

## [1] "Training time..."

training_time-start_time

## Time difference of 1.611486 mins

Question: How much time is taken by the model to train with 6 cores?

Cleaning

Remove unwanted data.

rm(numFolds)
rm(grid)
rm(model)

Question: Repeat the same to find out time required to train model with different number of cores.

caret parallel