Quick ML - Pocket Data Studio

Quick ML is a complete data-science studio for iPhone and iPad. Import real datasets, explore and clean them properly, train machine-learning models, and make predictions - the whole workflow, entirely on device. Built for data scientists and students of data science who don't always have a laptop with Python to hand.

FREE TO TRY, ONE PRICE TO OWN The free version is a real studio, not a demo: import files of any size, explore and clean without limits, train the classic algorithm for each task (linear regression, logistic regression, k-means, and the image classifier), evaluate honestly, and predict. When you want more, Quick ML Pro is a single purchase that unlocks all 19 algorithms, auto-tuning, imports from URLs, Kaggle, and PostgreSQL, refresh and lookup joins, dashboards, Core ML and Python export, project sharing, and applying anonymisation. No subscription, no account. Yours forever.

NO LAPTOP? NO PROBLEM On the train, in a lecture, at a client's site, on the sofa: open a dataset the moment you get it and have a trained, evaluated model before you're anywhere near a computer. A million-row CSV is fine - data streams into a local store and never loads into memory at once.

PRIVATE BY ARCHITECTURE Everything computes on your device. No accounts, no uploads, no analytics - your data never touches a server, which also makes it an easy answer when the dataset is sensitive. Projects sync privately between your own devices through your own iCloud.

IMPORT ANYTHING CSV, Excel workbooks, parquet, ZIP archives, URLs and JSON APIs, Kaggle datasets, even a PostgreSQL database. Sources can refresh later: re-import, append, or merge new rows on a key.

EXPLORE AND CLEAN PROPERLY Full descriptive statistics, histograms, box and violin plots, correlation heat maps, pair plots, and nine statistical tests with plain-English verdicts. A one-tap Data Health Check profiles every column and suggests cleaning steps, each with its reason. Every change is a recorded recipe step - fills, filters, tidy dates, dedupe, outlier trims, derived columns (date parts, bins, ratios, lags, rolling windows, text features) - replayable, synced, and individually undoable, like a pandas script you can swipe.

TRAIN REAL MODELS Regression, classification, and clustering: linear and logistic regression, decision trees, random forests, boosted trees, neural networks you design visually, k-nearest neighbours, SVM, Naive Bayes, k-means, DBSCAN, Gaussian mixtures. Honest seeded or time-ordered splits, cross-validated auto-tuning, class balancing, and full evaluation: R2 and RMSE, confusion matrices, ROC and AUC, permutation feature importance, learning curves.

STUDYING DATA SCIENCE? IT SHOWS ITS WORKING Every screen explains what the numbers mean and why each step matters, from quartiles to overfitting. Then How to Do This in Python exports your exact pipeline as a Jupyter notebook - every step as pandas and scikit-learn code with the reasoning in markdown

so what you did on the sofa becomes the coursework, and the concepts transfer straight to the tools you're learning.
WORKING DATA SCIENTIST? TAKE THE RESULTS WITH YOU Export any trained model as a Core ML .mlmodel and drop it into an Xcode project or Core ML pipeline. Run batch predictions over a whole file, score held-out validation sets, compare model versions, build live dashboards, and generate a client-ready PDF report of the entire project in one tap.

If you have data and a phone, you have a data-science workstation.