Online coding and Python classes in Cheltenham

The learner downloads Raikes's biography from Project Gutenberg, removes the lines that hold web addresses, lower-cases every word and lists each pair of neighbouring words. The book produces 71,376 different pairs, and 76.2 per cent of them appear only once. The first scoring rule is simply to count. The top four are "of the" at 1,340, "in the" at 700, "to the" at 530 and "miss beale" at 514. Only the last tells you anything about the book; the others would top the list for almost any English text.

The second rule is pointwise mutual information, which asks how much more often two words sit side by side than chance would predict. It rewards true partnerships, but it rewards rarity even more. Any pair seen exactly once, made of words that are themselves rare, gets the maximum score: 273 pairs tie at the top, including two editors' notes about an html version and the Internet Archive. The learner discovers that the scoring rule and the data cleaning are not separate jobs.

The third rule keeps the mutual information score but only counts pairs seen at least a set number of times. At 5, names dominate. At 20, the book finally speaks for itself: boarding houses, head mistress, lady principal, old pupil, all words from the working life of a school. One more trap turns up in testing. A tokeniser that only accepts the letters a to z splits Fräulein, which appears five times, into two meaningless pieces, so the program uses a pattern that accepts any letter in any alphabet.

Seeing the statistics behind the words is a strong reason for Cheltenham teenagers to keep learning to code in 2026. The longer argument is in why teenagers should still learn to code in 2026.