Python Core ML

July 2017 PyJax

Who am I

• David Fekke
• Chief Mobile Architect for Swyft Technology
• Formally ColdFusion/.NET/JAVA/PHP
• Founded JaxNode user group
• @jaxnode and @davidfekke
• New to Python

Python & Machine Learning

• Python and other languages like R and Julia becoming popular for Big Data analysis and Machine Learning
• Google introduced TensorFlow 2 years ago
• Other popular projects include Keras and Scikit

Use Cases

• Natural Language Processing
• Speech recognition
• Optical Character Recognition
• Computer Vision
• Not Hotdog!

Core ML

• Apple machine learning specification
• It is a model format
• Train models using a ML framework
• Convert models to Core ML format

Model Types

• Generalized Linear Models
• Pipeline Models
• Feature Engineering
• Support Vector Machines
• Tree Ensembles
• Neural Networks

Training Models

• Train your model with one of these frameworks
• Keras with TensorFlow
• Caffe
• XgBoost
• SciKit Learn
• libSVM

Convert Models

import coremltools
coreml_model = coremltools.converters.caffe.convert('my_caffe_model.caffemodel')

# Now save the resulting model in the Core ML model format.
coreml_model.save('my_model.mlmodel')

Core ML Models

• Open specification
• They have inputs and outputs
• They work like a function
• Simply drag and drop model in your Xcode project

Core ML advantages

• Does not require a data connection to run
• Can automatically decide whether to use the CPU or the GPU for processing
• A10 chip is as fast as 2.2GHz Core i7-5650U
• User Privacy
• Reduces Cloud costs

Core ML Disadvantages

• No Federated learning
• Can't retrain model

Demo

Questions?