tjwudi/uci_har_analysis

Course project of Getting and Cleaning Data Course

UCI_HAR_Analysis

This document describes steps used by run_analysis.R to extract tidy data from Human Activity Recognition Using Smartphones Data Set.

First, libraries are loaded, variables and helper functions are defined.

# Loading libraries
print("Loading libraries")
library('data.table', quietly = T)
library('reshape', quietly = T)

# Constants
uci.datasetUrl='https://d396qusza40orc.cloudfront.net/getdata%2Fprojectfiles%2FUCI%20HAR%20Dataset.zip'
uci.path=getwd()
uci.datasetPath=file.path(uci.path, 'data')

# Helpers
uci.helpers.readAsDataTable <- function (path) {
  data.table(read.table(path))
}

Download and extract dataset if we cannot locate the dataset folder.

if (!file.exists(uci.datasetPath)) {
  print("Downloading dataset");
  tmpfile <- tempfile()
  download.file(uci.datasetUrl, tmpfile, quiet = T)
  print("Extracting dataset")
  unzip(tmpfile, exdir=uci.datasetPath, junkpaths = T)
  unlink(tmpfile)
}

Load data using our helper function uci.helpers.readAsDataTable.

uciData.train.x <- uci.helpers.readAsDataTable(file.path(uci.datasetPath, 'X_train.txt'))
uciData.train.y <- uci.helpers.readAsDataTable(file.path(uci.datasetPath, 'y_train.txt'))
uciData.train.sub <- uci.helpers.readAsDataTable(file.path(uci.datasetPath, 'subject_train.txt'))
uciData.test.x  <- uci.helpers.readAsDataTable(file.path(uci.datasetPath, 'X_test.txt'))
uciData.test.y  <- uci.helpers.readAsDataTable(file.path(uci.datasetPath, 'y_test.txt'))
uciData.test.sub <- uci.helpers.readAsDataTable(file.path(uci.datasetPath, 'subject_test.txt'))
uciData.features <- uci.helpers.readAsDataTable(file.path(uci.datasetPath, 'features.txt'))

Set descriptive name for data we just loaded.

setnames(uciData.train.y, names(uciData.train.y), c("activityId"))
setnames(uciData.test.y, names(uciData.test.y), c("activityId"))
setnames(uciData.train.sub, names(uciData.train.sub), c("subject"))
setnames(uciData.test.sub, names(uciData.test.sub), c("subject"))

Merge train and test data by combining usages of rbind and cbind.

uciData.train <- cbind(uciData.train.x, uciData.train.y, uciData.train.sub)
uciData.test  <- cbind(uciData.test.x, uciData.test.y, uciData.test.sub)
uciData.all <- rbind(uciData.train, uciData.test)

(Optional) Destroy redundant big datasets.

remove(uciData.train)
remove(uciData.test)
remove(uciData.train.x)
remove(uciData.test.x)
remove(uciData.train.y)
remove(uciData.test.y)
remove(uciData.train.sub)
remove(uciData.test.sub)

Select measurements on mean and standard deviation.

setnames(uciData.features, names(uciData.features), c('featureId', 'featureName'))
featuresOnMeanOrStd <- uciData.features[grepl("mean\\(\\)|std\\(\\)", uciData.features$featureName),]
featureIdsOnMeanOrStd <- c(paste('V', featuresOnMeanOrStd$featureId, sep=''), 'activityId', 'subject')
uciData.all <- uciData.all[, featureIdsOnMeanOrStd, with = F]

Set descriptive names.

uciData.activityNames <- uci.helpers.readAsDataTable(file.path(uci.datasetPath, 'activity_labels.txt'))
setnames(uciData.activityNames, names(uciData.activityNames), c("activityId", "activityName"))
setnames(uciData.all, paste0('V',featuresOnMeanOrStd$featureId), as.character(featuresOnMeanOrStd$featureName))

Write tidy dataset to file.

write.table(uciData.all, file="dataset.txt", row.names = F)

To generate means for very variables on each subject and each activity, we first devide uciData.all into sub tables group by activityId and subject.

uciData.all$activityId <- factor(uciData.all$activityId)
uciData.all$subject <- factor(uciData.all$subject)
uciData.grouped <- split(uciData.all, list(uciData.all$activityId, uciData.all$subject))

Then calculate colMeans for each value in list uciData.grouped.

uciData.summaryMean <- list()
for (item in uciData.grouped) {
  item$subject <- as.numeric(item$subject)
  item$activityId <- as.numeric(item$activityId)
  uciData.summaryMean <- c(uciData.summaryMean, colMeans(item))
}
uciData.summaryMean <- matrix(
  unlist(uciData.summaryMean),
  ncol = nrow(featuresOnMeanOrStd) + 2,
  byrow = T)
colnames(uciData.summaryMean) <- colnames(uciData.grouped$`1.1`)

Finally, write this dataset to file as well.