Click to get started (first screen deliberately blank). Be aware that selecting zoom slider counts as a click so use + and - keys to zoom and cursor keys to move around the map.
Full Screen diffusion map showing related journeys. First we made a large square matrix indexed by stops in the transport network (approx. 12k): the value at cell (i,j) is the number of journeys between those two stops. We then multiply the number of journeys by the distance between the stops. This matrix is a representation of all the journeys travelled, weighted by how many people travelled and how far they went. One issue was reducing the data from raw tag-on and tag-off events to complete end-to-end journeys. Once we have that matrix we apply the diffusion map algorithm to identify the "most common journeys". It took a while to understand how to interpret the results, but you can think of hot spots on the map telling you that "people who went here also went there". For example, you see Femantle, UWA and the City all lit up in one slide. This says that people travelled between the three destinations in rough proportion to the intensity of the heat blobs on the map. Sometimes one spot is much brighter than others, with purple smudges in other locations. This represents something like a catchment area for a shopping centre or a university.
One avenue for further analysis would be to cluster the end-to-end journey dataset by card_id, trying to see explicitly which individuals went where. It would take some experimentation to find the best representation for that analysis as there would be an awful lot of data to draw.