AI and programming classes in Newtownards

The learner downloads NISRA's Census 2021 table MS-A02, which counts residents in five-year age bands for every settlement, and keeps the 146 settlements with at least 1,000 people. Each becomes a list of 19 percentages, from 0 to 4 years up to 90 and over. In those 19 dimensions, the towns whose age mix is closest to Newtownards are Enniskillen, Coleraine, Ballymoney, Ballymena and Larne. Python then flattens everything to two dimensions twice: once with principal component analysis (PCA), a straight projection, and once with t-SNE, which tries hard to keep each town next to its closest matches.

Even the most faithful map keeps only about half of each town's ten real nearest neighbours; the other half are strangers placed close by accident of the squashing. t-SNE does a little better than PCA at local neighbourhoods, and scores slightly higher on trustworthiness, a standard check of whether map neighbours are real ones, but its picture changes with the perplexity setting and with the random seed. Across nine t-SNE runs, the five towns drawn nearest to Newtownards changed each time, with Limavady, Saintfield, Randalstown and Killyleagh all appearing, while Ballymena, one of its five true closest matches, never appeared at all. PCA, meanwhile, explains 55.0% and 11.8% of the variation along its two axes, so a third of the information is simply not on the page.

A Newtownards teenager who reruns a pretty plot before believing it will not be fooled by an AI-generated one, and that habit starts with writing code. The longer argument is in why teenagers should still learn to code in 2026.