Changelog

What's changed

Fit Lab is a draft, built in the open with its author. Newest first.

0.3 · 3 October 2026

Try it on real wells: every chapter ends with its lesson on four North Sea wells from the FORCE 2020 competition, and a Data page describes both datasets.
The workbench: every algorithm, both datasets and every setting on one screen, with the setup kept in the address so it can be shared.
Challenges: five beat-the-algorithm puzzles, each opening once you've read its chapter, with your best kept in this browser.
A dark variant, with its own colours rather than inverted ones, and a settings menu (Aa) for the theme and the text size, from 90% to 130%.
Metrics: fill in the confusion matrix yourself, a tap at a time, and watch every score move.
Glossary, Assumptions, About, Creator and Changelog pages.
Chapters you've read are ticked on the contents page.
Words you'll meet opens in a box from the bottom right of every chapter, with a filter, so the terms are to hand while you read. Chapter 5's box carries its cheat sheet.
Rights: © 2026 Andy McDonald, all rights reserved, as across the Labs, with third-party notices for the fonts and the FORCE 2020 data.
Fixed: links could lead to the contents page at the wrong address, and from there every link went astray.
Fixed: the glossary's lists sat at the left edge of the page, and four chapters still said the real wells were coming.

Chapter 8, neural networks: draw the sonic yourself, watch a small network invent its own logs as it trains, see when to stop, and test five networks on a blind well.
Chapter 9, isolation forest: flag the washout yourself, watch random cuts isolate strange samples, and see what the forest really flags across six wells.
Chapter 0, a log is a table: a log becoming a table and then a crossplot, finding the coal, and features and labels.
Random forests: out of bag explained, with its catch, and the terms that come with forests.
Metrics: ROC and precision–recall curves, each facies against the rest, and the classification report.
A "Words you'll meet" glossary in every algorithm chapter.
Fixed: a huge logo flashed up on changing page; k-means' starts in scikit-learn were described wrongly.

First build: the six-well teaching field, and chapters 1 to 7 (k-means, linear regression, k-nearest neighbours, train, validate and test, metrics, decision trees and random forests).