Core_ML_in_Depth.md

WWDC 2017

Table of Contents

• Core ML in Depth - Thursday
  • Sentimental Analysis
  • Predictive Keyboard
    • Hardware Optimized
• Deploying Core ML Models
  • Core ML Tools
    • Conversion Workflow
    • What's CoreML Tools
  • Conversion Demo
  • Summary

Core ML in Depth - Thursday

Session video and resources: https://developer.apple.com/videos/play/wwdc2017/710/

Sentimental Analysis

• Use case: Sentiment analysis app which gives mood output according to the text input
• Use NLP (NSLinguisticTagger) -> Get Word count and tokenize -> Feed it to the ML model

Predictive Keyboard

• Task: Next word prediction
• Sequence of words -> Model -> Next word choices & state
  • Next word & state is fed into the model recursively

Hardware Optimized

• Core ML uses Accelerate on top of CPU and Metal on top of GPU

Deploying Core ML Models

• Example models under https://developer.apple.com/machine-learning

Core ML Tools

• Fully open sourced

Conversion Workflow

• Model Source (e.g Caffe) -> Xcode -> ML Model & Swift ->
• pip install coremltools

What's CoreML Tools

• Convert from other formats
• Build your own converter
• Compatible & Extensible

Conversion Demo

Sample python code

import coremltools
caffe_model = ('flowers.caffemodel', 'flowers.prototxt')
labels = 'labels.txt'

coreml_model = coremltools.converters.caffe.convert(caffe_model, class_labels=labels, image_input_names='data')
coreml_model

// Output of dictionaries with the results

coreml_model.save('FlowerPredictor.mlmodel')

Summary

• Easy integration of ML models
• Rich datatype support
• Hardware optimized
• Compatible with popular formats