Online coding and Python classes in Larne

The learner downloads OpenStreetMap data for a rectangle over Larne in twelve tiles: 65,602 map elements, from single points to whole roads and boundaries. Each element records who last edited it and in which batch of edits, called a changeset. Counted exactly with a Python set, the stream holds 216 different contributors and 1,128 different changesets. Keeping every element ID in a set took about 5.5 MB of memory. HyperLogLog replaces the set with a fixed row of small counters, called registers, and the learner tries four sizes, each with 200 different hash functions to see how much the answer varies.

Each fourfold increase in registers roughly halves the error, as the textbook formula of 1.04 divided by the square root of the number of registers says it should. With 1,024 registers, about 768 bytes, the estimates for all three counts land close: a typical error of 1.9% for the 65,602 element IDs, 2.0% for the changesets and 1.6% for the 216 contributors. The memory stays the same however many items flow through; the exact set grows with every new one. The price is a small, predictable error, and the learner has to decide whether that price is acceptable for the job.

Knowing when an estimate is safe lets Larne teenagers use data and AI with judgement, and learning Python is where that judgement grows in 2026. The longer argument is in why teenagers should still learn to code in 2026.