Online coding and Python classes in Musselburgh

From OpenStreetMap the learner keeps the public roads in a rectangle over Musselburgh, 120.5 km of them, and merges points where a road merely bends, leaving 1,124 junctions joined by 1,286 road segments. A 1,124 by 1,124 table starts with the length of each direct segment and infinity everywhere else. Floyd-Warshall then makes 1,124 passes: on pass k, any pair whose route would be shorter by going through junction k gets its entry lowered. That is about 1.42 billion small updates, which numpy handles in 10.1 seconds on an ordinary laptop by updating a whole row block at a time.

To be sure the table is right, the learner compares it with Dijkstra's algorithm run from a sample of junctions: every value agrees. Dijkstra from every junction would take an estimated 4.2 seconds here, so on a sparse road network the repeated single-source approach is faster, while Floyd-Warshall wins on simplicity and on dense networks. Across all pairs, the average drive between two junctions is 3.21 km and the longest shortest route, the network's diameter within the rectangle, is 10.26 km.

Most pairs of places are about one and a half times as far by road as in a straight line, but Inveresk to Monktonhall stands out at nearly three times. The table cannot say why; a learner would look at the map to see what the roads must go round. Distances here use public roads inside the rectangle only, so any quicker route leaving the box is not counted.

Being able to weigh an algorithm's cost lets Musselburgh teenagers judge AI-written code instead of simply running it, which is reason enough to learn Python in 2026. The longer argument is in why teenagers should still learn to code in 2026.