createml.md

Introducing Create ML

• Focus on images, text and tabular data
• Folders as labels, images as data for that label
• Transfer learning is used in Create ML by using an existing image classifier, augmented with your data
• This is a real quick way of training
• Playground is used to provide a UI for machine learning
• MLModelImageClassifierBuilder
• Available on Mojave

Doing it with code

• You can provide certain parameters like a source and test directory
• Playground REPL mode
• Model can be written to a file
• MLImageClassifier, model.evaluation
• You can run .swift files on the command line

Natural Language

• Data source can be folder with txt files
• JSON and CSV are also supported

Tabular data

• Uses MLDataTable
• CSV, JSON and Code are supported
• data tables can be filtered and more
• MLLinearRegression, give it data and the target column
• Can also do MLBoostedTreeRegression
• MLRegressor will pick the best algo for you