datadrover/tidydata

Creates tidy summary data set from mobile sensor data

Repository contents

1. README.md (this file)
2. CodeBook.md : code book for data-set
3. run_analysis.R : R script for data tidying and transformation
4. tidymeans.csv : csv output of transformed and aggregated data (means)

Introduction

run_analysis.R is an R script that extracts the required mean and standard deviation measurements from a specific study data gathered with Samsung Galaxy smartphones. Data are recombined into a unified set along with subject (participant) reference numbers and activity labels. Text labels are appropriately modified by the script. Users may want to refer to the code book CodeBook.md in this repository and the original description files which are listed below.

Assumptions

The script has been tested under R 3.02 and R 3.1. It requires the Reshape2 package, which can be be installed from a cran mirror. The script assumes that the data has been downloaded from: https://d396qusza40orc.cloudfront.net/getdata%2Fprojectfiles%2FUCI%20HAR%20Dataset.zip

As required by the specification, the data should be unzipped and R's working directory set to the base directory (i.e: UCR\ HAR\ Dataset).

Data

The original data is prepared as supervised learning set. It is labelled in terms of six activities and divided into training and test sets.Training and test sets are contained in corespondingly named subdirectories. All of the original data is contained in text files. Columns are space separated.

The following files from the original data are used:

1. ./train/X_train.txt (sensor data)
2. ./train/Y_train.txt (activity markers)
3. ./train/subject_train.txt (subject numbers)
4. ./test/X_test.txt (sensor data)
5. ./test/Y_test.txt (activity markers)
6. ./test/subject_test.txt (subject numbers)
7. ./activity_labels.txt (the six activity labels)
8. ./features.txt (list of the 561 features in the X files)

See these description files in the original zip:

1. ./README.txt
2. ./features_info.txt

Processing

For the training and then the test set: the sensor data, activity markers and subjects numbers are read in from the corresponding text files. The 561 sensor features are reduced to the 66 features containing mean and standard deviation data. Mean frequencies (weighted average of the frequency components have been purposefully excluded,but can be included by a minor modification indicated in the code comments.

The column names are cleaned up according to the "house style": Names are converted to lower case and illegal characters removed. With the exception of "gravity", all features describe body movements. Therefore "Body" items (including duplicates) are removed. The rational for the adopted house style is that camel case does not help with comprehension in this case, but may lead to typing errors. The "t" and "f" prefixes denote time and frequency domains respectively. Given that names are already long, they have not been expanded.

The activity labels have been similarly modified. For semantic correctness, "LAYING" has been changed to "lyingdown".

Output

The output is a tidy data set with 68 variables and 180 observations. (30 subjects by 6 activities), The variables are required means for the relevant features preceded by the subect number and activity label. Thus, it is possible to look up mean figures by subject and activity. Arguably, this leads to a rather unwieldy look-up table, but there is no justifiable basis for horizontal aggregation. The output meets the the criteria for tidy data as defined by Wickham (see references):

1. Each variable forms a column.
2. Each observation forms a row.
3. Each type of observational unit forms a table.

References

https://github.com/hadley/tidy-data