The goal of this project is to prepare tidy dataset from Human Activity Recognition Using Smartphones Data Set.

Project consists of these files:

• README.md - file you are reading now
• run_analysis.R - R script with functions to prepare tidy dataset
• CodeBook.md - a code book that describes the variables, the data, and transformations performed to clean up the data

To prepare a tidy dataset perform the following steps:

1. Start R console.

2. Set working directory to the project dir with script file run_analysis.R.

3. Load functions from R script:

    > source('run_analysis.R')
   

4. Run main function:

    > tidy_har_dataset(dest)
   

dest is file name where tidy dataset will be outputed.

The main function tidy_har_dataset() does not assume that UCI HAR Dataset/ directory is in the working dir. However, if it is already present, data from it are used.

If directory is not present, function looks for uci_har_dataset.zip file. If this file is also not present, data are downloaded from the original URL and saved into uci_har_dataset.zip. Then zip file is extracted.

After making sure that data dir is in place, the main data transformation proceeds in the following steps:

1. Test and train data are read from UCI HAR Dataset/test/ and UCI HAR Dataset/train/ dirs respectively and combined into one dataset.

   For both of test and train data there are three files: X_...txt, y_...txt, subject_...txt that need to be read and combined. These three files have the same number of lines corresponding to the same observations. subject_...txt provide subject identifiers for each observation, y_...txt provides activity codes for each observation, and X_...txt provides measurement data for each observation. Data from these three files are column-binded to yield a dataset with column Subject, Activity and 561 columns for measurements with automatically constructed column names V1, ..., V561.

   Loaded test data consist of 2947 rows-observations, while train data consists of 7352 rows-observations.

   Finally, these two datasets are row-binded, resulting in a merged dataset of 10299 rows corresponding to observations and 563 columns (Subject, Activity, and 561 measurements).

2. Only columns with measurements on the mean and standard deviation for each measurement are selected (obviously, together with Subject and Activity).

   File UCI HAR Dataset/features.txt lists all the features that were measured during the experiment. It has 561 lines with measurement names corresponding to 561 columns V1, ..., V561 in the data that were loaded and merged in the first step.

   Measurements of mean and standard deviation for each signal have -mean() and -std() in their names. I do not include measurements with -meanFreq() and the last 7 with Mean in their names, as they have a rather different meaning.

   To get names of columns to be extracted, function finds row indexes for features with matching names (using grep), transforms them into corresponding names in the form V? and then uses select_ function from dplyr package to select only those columns. Activity and Subject columns are selected as well.

   After this step only 68 columns (66 measurements) are left.

3. Activities are descriptively named in the main dataset.

   Up until now activities are represented by their codes in Activity column that are not descriptive at all.

   Similarly as with features, there is a file UCI HAR Dataset/activity_labels.txt that lists activities names with their corresponding codes. The data from this file are read first. Then activity names are transformed to be more readable: they are lowercased and the underscore is replaced with a simple space.

   Finally, activity codes from Activity column in the main dataset are replaced with corresponding activity names from activity data set.

4. Measurement columns are descriptively named.

   Up until now measurement columns still have their automatic column names in format V?. In this step, they are renamed to more descriptive names using feature-measurement names that were loaded and used in the 2nd step.

   Straight features names could be used but I transform them to be a little bit more tidy and reader-friendly: symbols ()- are removed, names are camel-cased, also some errors are fixed, like BodyBody in certain variables. Thus column corresponding to measurement fBodyBodyGyroJerkMag-mean() is named fBodyGyroJerkMagMean.

5. A new tidy dataset is created with the average of each variable for each activity and each subject. I selected long format of tidy data.

   Previously prepared dataset is melted so that columns corresponding to measurements become values in new column Variable and value of measurement goes to a new column automatically named value.

   Finally, data are summarized, grouping by Subject, Activity and Variable columns, and creating new column Average with mean number of values in each group.

   This final dataset contains 11880 lines - one for each combination of 30 subjects, 6 activities and 66 measurements.

   If dest argument was provided to the function, final tidy dataset is outputted to it.