Coding and AI classes in Portadown

The learner downloads OpenStreetMap data for a rectangle around Portadown and keeps every way a pedestrian may use. Joining points where a path merely bends leaves 4,504 junctions and dead ends linked by 5,625 segments. Degree is easy to count: the number of streets meeting at each junction. Closeness needs the walking distance from every junction to every other, over 20 million pairs, which Python computes with Dijkstra's algorithm on a sparse matrix; a junction's closeness is one over its average distance to all the rest.

Closeness finds the town centre with surprising accuracy: the junction with the shortest average walk to everywhere else sits 91 m from the point OpenStreetMap uses to label Portadown, and half the network lies within 1,659 m of it. Degree tells a different story. The twenty most central junctions all have just three or four streets, while the six junctions where five or more streets meet have a median closeness rank of 1,904 out of 4,504. A rank correlation of 0.205 means that knowing how many streets meet somewhere tells you very little about how central it is.

One caution: closeness also rises with distance from the edge of our rectangle (correlation 0.777), partly because the town centre really is in the middle and partly because cutting the map makes edge junctions look remote. Moving the rectangle would shift the scores near its borders, which is why the learner reruns it before trusting the details.

A Portadown teenager who asks which definition sits behind a ranking will not take an AI league table on trust, and writing the code yourself is the quickest way to learn that. The longer argument is in why teenagers should still learn to code in 2026.