A Guide to Turi Create

WWDC 2018

What is Turi Create?

https://github.com/apple/turicreate

• Python library for creating Core ML models
• Easy to use, no need to be an ML expert
• Task focused APIs
• Cross platform (Mac, Linux)
• Open source

5 step recipe for creating Core ML models

Step 1: Task

Step 2: Data

What is an SFrame?

• Disk backed, tabular data struct
• Common data manipulation tasks
• Work with text, images, and json
• Interactively explore and visualize data

Step 3: Model

• Model creation customized to task
• State of the art
• Small or large amounts of data

Step 4: Evaluate

Step 5: Deployment

Demo

1. Loaded images and annotations into the SFrame

2. Interactively explored data

3. Created a model

4. Evaluated quantitatively

5. Exported to Core ML

Turi Create 5.0

• New Task
  - Style Transfer
• Performance
  - Native GPU Acceleration on Mac
• New Deployments
  - Recommenders
  - Vision Feature Print powered models

Style Transfer

Demo

1. Loaded images into the SFrame
2. Created a model

3. Stylized images

4. Visualized predictions
5. Exported to Core ML

Mac GPU acceleration

Personalization

• Task: Recommend items for users
• Data: Historical preferences
• Deployment:
  - Core ML custom models
  - macOS 10.14, iOS 12
• Top community feature request