If you can see a table, Quick ML can now import it. - Word and PDF documents: tables come in with merged cells and multi-page layouts handled. - A ZIP of same-shaped files combines into one dataset with a source-file column. - Keep History: a refresh mode that appends every fetch with its timestamp, turning a live page into a time series. - Tidy Durations and Tidy Text, and Tidy Numbers now reads percent signs, currency and bracketed negatives. - Calculated columns, eighteen of them: date difference and date shift, compare columns, powers, roots and functions, split, join, duplicate, extract §a number from text, standardise, running total and rank, and more. - Edit or remove a single row, recorded as a recipe step like everything else. - Various fixes and improvements throughout, including honest test scores for image classifiers.

Quick ML can now import from almost anywhere on the web: - Web pages: point at a page with a table on it, tap the table you want, and it becomes a dataset you can clean, train on, and refresh later. Handles multi-row headers, image cells, and results split across numbered pages or next links. - JSON and XML APIs: fetch a sample, then shape your table by tapping it - pick fields, expand lists into rows, set up pagination and sign-in once. Every refresh replays your exact steps. - GraphQL: post a query and map the reply the same way, cursor paging included. - Local JSON and XML files import through the same visual mapper, free. Also in this update: - Combine two projects into one dataset on a shared key column, free. - Extract columns from JSON packed inside a text column, free. - Sharing a project now carries the whole import and cleaning recipe: the person you send it to reconnects the source with one tap and gets the same cleaned table you had. - How to Do This in Python now exports the data source too, so the notebook fetches the same data your project does. - Select several projects at once to delete them. - Fixed a crash when opening a shared project file while the app was closed.

Take more control of your training with new settings: - Ridge (L2) and lasso (L1) regularisation for linear and logistic regression, to rein in overfitting or let the model drop weak features on its own. - Learning rate, min child weight, and row and column sampling for boosted trees, plus row and column sampling for random forests. - Auto-Tune now searches regularisation for linear and logistic regression too. Also in this update: - Model page charts now carry titles, and models that share a name chart separately in Compare Models. - Clustering models are named after the algorithm you actually chose. - Better number entry everywhere: the right keyboard for each field, tap anywhere on a row to edit its value, and typing replaces the old number instead of adding to it.

Quick ML is now free to try. Import data of any size, explore and clean it, train the classic algorithm for each task, evaluate, and predict, all free. One purchase of Quick ML Pro unlocks every algorithm, every data source, dashboards, and every export, forever. No subscription. Bought Quick ML as a paid app? Pro is already yours and unlocks automatically. Also in this update: Big performance improvements for very large image datasets. Rate Quick ML from the About screen, plus an occasional rating prompt after successful training.