This data archive for this analysis was first downloaded by this author on 10/31/16, but it is also downloaded from a Cloudfront server [(here)] (https://d396qusza40orc.cloudfront.net/getdata%2Fprojectfiles%2FUCI%20HAR%20Dataset.zip), by the script run_analysis.R in this repo. This data was originally obtained from a UCI Website [(here)] (http://archive.ics.uci.edu/ml/datasets/Human+Activity+Recognition+Using+Smartphones).
The results produced by the analysis script are written to disk as two data files:
tidy_data.txt, which contains selected data from the original UCI analysis itself, in space delimited formattidy_summary.txt, which contains summary metrics of the feature data from UCI, calculated byrun_analysis.R
To produce the dataset from which (1) was selected, investigators used Samsung cell phones to record gyroscopic and accelerometer sensor readings of 3-axial linear acceleration and 3-axial angular velocity at a constant rate of 50Hz. The cell phones were worn on the belts of 30 subjects while they engaged in six, video-recorded acitvities (see list below). From the sensor readings, 561 different features were calculated, of which 66 were selected for this analysis. Each observation of 66 features corresponds to a row in (1), calculated based on 128 readings taken during a 2.56 sec time window. The 66 features selected for this analysis correspond to signal means and standard deviations: original feature names containing mean() or std(). Features not selected for this analysis include max(), min(), correlation(), and a diverse set of other caclulated variables.
Each row of feature values in (1), and in the full dataset in the file features.txt of the UCI archive, was calculated based on readings taken while a specific subject engaged in a specific activity. Column 1, numbered 1 to 30, contains the subject ID number, and is encoded as a factor variable in the loaded dataframes. Each row of column 2 takes on one of the six values below, and is also encoded as a factor variable. The values below are transformed from those in the original dataset by lower-casing and substitution of a "." for hyphens between words.
laying
sitting
standing
walking
walking.upstairs
walking.downstairs
Columns 3 through 68 of the files contain the original feature data (file 1) and summary metrics (file 2). In the data of file (1), all features were normalized and bounded within [-1,1]. This data consists of 10,299 rows, each corresponding to data collection over a specific 2.56 sec window for a specific subject-activity pair. The summary metrics in file (2) are averages across all time windows (rows of 1) associated with a given subject-activity pair, resulting in just 180 rows (6 activities * 30 subjects).
The names of the variables selected for the dataset in (1) are listed below alongside their column number in files (1) and (2). Note that t is used to denote time domain, and f is used to denote frequency domain; acc, gyro stand for acceleration and gyroscope. The terms body and gravity refer to a separation of acceleration into body and gravity components. The magnitudes mag are calculated as Euclidean norms of specific signals by taking into account variation over all three axial directions x, y, and z. The term jerk refers to derived signals over time of body linear acceleration and angular velocity. For more detail on the pre-processing of raw sensor readings (summarized here), see the files features_info.txt and README.txt, both included in the UCI archive for the experiment.
The variable names shown here are transformed for readability from those used in features.txt in the original data archive. All letters were lower-cased; all non-alphanumerics, white-space, and variable numbers were removed; and a dot (".") was placed between each pair of word fragments. For the averages in file (2), the variable names below are prefixed with the characters avg.
[3] "t.body.acc.mean.x" "t.body.acc.mean.y" "t.body.acc.mean.z"
[6] "t.body.acc.std.x" "t.body.acc.std.y" "t.body.acc.std.z"
[9] "t.gravity.acc.mean.x" "t.gravity.acc.mean.y" "t.gravity.acc.mean.z"
[12] "t.gravity.acc.std.x" "t.gravity.acc.std.y" "t.gravity.acc.std.z"
[15] "t.body.acc.jerk.mean.x" "t.body.acc.jerk.mean.y" "t.body.acc.jerk.mean.z"
[18] "t.body.acc.jerk.std.x" "t.body.acc.jerk.std.y" "t.body.acc.jerk.std.z"
[21] "t.body.gyro.mean.x" "t.body.gyro.mean.y" "t.body.gyro.mean.z"
[24] "t.body.gyro.std.x" "t.body.gyro.std.y" "t.body.gyro.std.z"
[27] "t.body.gyro.jerk.mean.x" "t.body.gyro.jerk.mean.y" "t.body.gyro.jerk.mean.z"
[30] "t.body.gyro.jerk.std.x" "t.body.gyro.jerk.std.y" "t.body.gyro.jerk.std.z"
[33] "t.body.acc.mag.mean" "t.body.acc.mag.std" "t.gravity.acc.mag.mean"
[36] "t.gravity.acc.mag.std" "t.body.acc.jerk.mag.mean" "t.body.acc.jerk.mag.std"
[39] "t.body.gyro.mag.mean" "t.body.gyro.mag.std" "t.body.gyro.jerk.mag.mean"
[42] "t.body.gyro.jerk.mag.std" "f.body.acc.mean.x" "f.body.acc.mean.y"
[45] "f.body.acc.mean.z" "f.body.acc.std.x" "f.body.acc.std.y"
[48] "f.body.acc.std.z" "f.body.acc.jerk.mean.x" "f.body.acc.jerk.mean.y"
[51] "f.body.acc.jerk.mean.z" "f.body.acc.jerk.std.x" "f.body.acc.jerk.std.y"
[54] "f.body.acc.jerk.std.z" "f.body.gyro.mean.x" "f.body.gyro.mean.y"
[57] "f.body.gyro.mean.z" "f.body.gyro.std.x" "f.body.gyro.std.y"
[60] "f.body.gyro.std.z" "f.body.acc.mag.mean" "f.body.acc.mag.std"
[63] "f.body.body.acc.jerk.mag.mean" "f.body.body.acc.jerk.mag.std" "f.body.body.gyro.mag.mean"
[66] "f.body.body.gyro.mag.std" "f.body.body.gyro.jerk.mag.mean" "f.body.body.gyro.jerk.mag.std"