Skip to contents

Data

data.Rmd

This vignette describes the data included in the package, how it was prepared, and why it was prepared in the way it was. It does not describe the final form of the data as it is included in the package in detail (see the data description for that, e.g. ?tracks), but rather how the original raw data looked, and how and why it was pre-processed to end up the way it is stored in the data-raw folder (see package sources on Github).

Acquisition

The raw data was captured using a (now discontinued) bike computer of the model “Sigma ROX GPS 11.0”, with an additional chest strap heart rate sensor, and a combined speed and cadence sensor (both sensors were bought bundled with the bike computer itself).

The data was exported using the (proprietary) Sigma software, and then exported as CSV files.

The data set comprises 157 trip recordings (“tracks”), each of which is a sequence of data points taken at a five second interval. These data points were then aggregated into one summarized row per track. Additionally, tracks were manually classified by route, and direction in which the route was driven. These three data sets are available in the package under the following names:

  • track_details contains the recorded data point sequences of all tracks (approx. 160k rows)
  • tracks contains the summarized track information, one row per track (157 rows)
  • track_classes contains the classification information for each track, i.e. a class ID and direction (157 rows)

Tracks are identified by date, thus if, for example, summaries in tracks should be extended with class labels for each track, tracks can be joined with track_classes by the date key.

For convenience, there are two additional data sets included (swiss_cantons and swiss_lakes), containing geographic data for the cantons and lakes of Switzerland. These can be used to plot a geographic frame of reference when plotting tracks. Since the trip routes are all within the canton of Berne, the canton data with a KTNR (“Kantonsnummer”) of 10 (i.e. Berne) is probably the most useful. The source for the geo data is the Swiss Federal Statistical Office (Bundesamt für Statistik, BfS), where the raw data can be downloaded from this page.

Cleaning

The data available in the package is a cleaned-up version of the raw data available in the package sources. Cleanups performed were:

  • Some 0/0 GPS coordinates were set to NA values. These values were probably caused by some sort of GPS hiccup, so to clarify that there is no real data available for these points, and to prevent accidentally using the invalid coordinates for computations, they were set to NA.
  • Some duplicate data points were removed. There were some data points in one track sequence that had the same TrainingTimeAbsolute values. This looked like a temporary freeze of the bike computer, and there were some obviously-wrong values for distance and altitude changes in those rows, so only the first of those rows was kept, the others were completely removed.
  • Power zone data was removed. The raw data contained columns for power zones, however, there is no power sensor installed or calibrated, so the raw data in these columns is bogus. To prevent confusion, this data was not included in the package data, i.e. power zone-related columns were removed.

The uncleaned data is still available in the raw CSV files available in the package sources, and the tracks.R file in the data-raw folder contains the code that was used to perform the cleanup.

Geographic Censoring

Since the detailed track data points contain GPS data, the start and end points of the tracks would show the exact starting points for the trips, i.e. my address. Therefore, data points at the start and end of a trip, and within a certain region, were removed. Additionally, the tracks were then adapted, so that the first point of the censored track has a time and distance of 0, so that deducing the true starting point of a track based on the censored data points is not possible (or at least more difficult).

This approach of just cutting off parts of the track and resetting the first point to 0 is slightly distorting the data, but this was deemed acceptable. The initial idea of removing the first and last points few, but not resetting the time and distance of the first uncensored point and instead adding random noise to censor individual tracks but keeping the average track time and distance somewhat more intact, was not pursued further because it is a more complex approach, and the risk of accidentally leaking data was deemed higher.

The raw data for the tracks is only available in this censored form, i.e. the CSV files in data-raw/tracks/censored are the tracks with a trimmed start and end. The original, unmodified raw CSV files are not publicly available. The exact details of the censoring procedure can be found in the data-raw/tracks.R file.

Categorizing

The tracks are not all along different routes, some routes were driven multiple times (some much more often than others). The tracks were (manually) classified by their routes, with arbitrary numeric identifiers for the routes. Routes that were only driven once did not get their own class, thus some tracks have a route class of NA. Furthermore, one route was driven in different directions, tracks along that route all have the same class, but are additionally labelled as clockwise or counter-clockwise. All other routes were only ever driven in one direction, these tracks are not labelled for direction (i.e. their direction column contains NA).

Classification was done visually, by plotting the tracks, storing the plots as image files, and then sorting them into folders. The commented code that was used for classification can be found in data-raw/track_classification.R.