This week

After 2 months we finally have the dataset from Nokia.

The diary has not been updated for all this time because without a dataset there was nothing to do, I spent the last month writing the thesis.

Google Colaboratory

The dataset contains data of 191 unique user.

From a first exploration of the dataset it seems that there are three possible ways:

  1. Use the visits_10min table to have sequence of visited places
  2. User GPS and WLAN information to discover relevant places (e.g. using PlaceSense) and derive from them a sequence of visited places
  3. Use GPS and WLAN information as coordinates to then subdivide the space into grids to have a coarse granularity

The Nokia LDCC paper says that:

<aside> 💡 The fraction of time where participants are collecting detailed GPS data is 2.4 percent. Besides GPS, less accurate localization techniques can be used, such as WLAN triangulation. By combining GPS data and WLAN scans, we have been able to estimate the location of 57 percent of the access points.

Participants are covered by such access points 61 percent of the time, which allows a rough localization. The accuracy of this method depends of course heavily on the density and placement of the access points.

For the remaining fraction of the time, users can simply be positioned using the cell ID information. Note that a shortcoming of the current software is that it reports the ID and received power of only one cell tower at a time, which precludes GSM triangulation. Figure 4 illustrates the time

</aside>