Modeling journey time on the London tube network using graph representation learning.
To begin work on this, I'm using a toy dataset for the following subset of the tube network (the stations not greyed out).
- The
tubeulator
library is used for data download- It reads the TfL API to get names and lat./long. of stations
- Stations serving multiple lines are duplicated by defining a station as line-specific
- When encoding as a node feature, we use a single number to indicate the line
- We represent line changes using a transfer edge between the different lines at the same station
- (WIP)