Turi Create is an Apple-maintained machine learning framework written in C++ that abstracts away algorithmic complexity to enable developers to build custom ML models without requiring deep expertise in machine learning. The framework is designed specifically to simplify the development pipeline from data exploration through model deployment, with particular emphasis on integration with Apple's Core ML ecosystem for iOS, macOS, watchOS, and tvOS applications.
The framework supports a comprehensive range of machine learning tasks across multiple data modalities. Users can build recommender systems for personalized recommendations, image classification models for labeling images, drawing classifiers for recognizing pencil and touch gestures, sound classifiers for audio analysis, and object detection systems including one-shot object detection that learns from single examples. Additional capabilities include style transfer for image stylization, activity classification using sensor data, image similarity matching, general-purpose classifiers and regression models, clustering for grouping similar data points, and text classifiers for sentiment analysis. This breadth of functionality positions Turi Create as a multi-domain machine learning toolkit rather than a specialized tool.
The framework prioritizes accessibility through an emphasis on ease of use, allowing developers to focus on tasks rather than underlying algorithms. Built-in streaming visualizations enable exploratory data analysis, and the system supports diverse input types including text, images, audio, video, and sensor data. Turi Create is designed to handle large datasets on single machines through optimized performance, and models export directly to Core ML format for seamless deployment in Apple applications.
Platform support spans macOS 10.12 and later, Linux with glibc 2.10 or higher, and Windows 10 via WSL. The framework requires Python 2.7 through 3.8, x86_64 architecture, and a minimum of 4 GB RAM. While GPU acceleration is optional, certain models including image classification, object detection, one-shot object detection, style transfer, sound classification, drawing classification, activity classification, and image similarity can achieve 9-13x speedup when GPU resources are available. GPU support is automatic on macOS, with additional configuration required for Linux systems.
According to GitGenius activity tracking, the repository shows extended response latencies with a median issue and pull request response time of approximately 47,478 hours and a mean of 47,162 hours across tracked items. The most frequently applied issue labels are toolkits with three occurrences, followed by priority labels p1 and p3 with two occurrences each. The repository shares contributors with microsoft/vscode, microsoft/typescript, and rust-lang/rust, indicating cross-pollination with major open-source projects in developer tooling and systems programming.
