fcanas/bayes

Naive Bayes Classifier in Swift for Mac and iOS

Bayes

Bayes is a Naive Bayes Classifier for iOS and Mac platforms.

Bayes is implemented in Swift and takes advantage of generics to enable any Hashable, Equatable type of your choosing or creation for use as category or feature.

Installation

Swift Package

Bayes is built as a Swift package.

CocoaPods

Bayes is available as a CocoaPod, but has not yet been versioned and submitted to trunk. So to use it:

pod 'Bayes', git: 'https://github.com/fcanas/Bayes.git'

Since Bayes is written in Swift, you will need to be using a recent version of CocoaPods (>0.36) and you may need to add use_frameworks! to your Podfile. See this blog post for more information.

Framework

The project includes framework targets for iOS and Mac. Pull the repository and build the appropriate target for you. When Bayes reaches a release milestone, it will be appropriately tagged and should be easily available via Carthage.

Use

var eventSpace = EventSpace<String, String>()

eventSpace.observe("Cat", features: ["paw", "tail", "claw"])
eventSpace.observe("Cat", features: ["stripe", "tail", "whisker", "ear"])
eventSpace.observe("Cat", features: ["meow", "vertical pupil"])

eventSpace.observe("Dog", features: ["paw", "tail", "bark"])
eventSpace.observe("Dog", features: ["wag", "fetch", "tail", "paw"])

var classifier = BayesianClassifier(eventSpace: eventSpace)

XCTAssertEqual(classifier.classify(["claw", "tail"])!, "Cat", "Should categorize as Cat, due to claw")
XCTAssertEqual(classifier.classify(["bark", "tail"])!, "Dog", "Should categorize as Dog, due to bark")
XCTAssertEqual(classifier.classify(["tail"])!, "Cat", "Should categorize as Cat, due to base rate")
XCTAssertEqual(classifier.classify(["paw", "tail"])!, "Dog", "Should categorize as Dog, due to prevalence of paw")

Contribute

Contributions are welcome. Please note the following guidelines

• Test public functionality
• Make everything as private as possible; use private and internal aggressively and only expose what is necessary for external functionality
• Favor structs over classes
• Extract standard mathematical functions that may simplify implementation of future models, e.g. product, argmax

bayes's Issues

Gaussian Event Model

Modeling a continuous event space requires a new event model. I think this warrants creating an event space protocol.

Multinomial Event Model

I think this warrants creating an event space protocol.

see #2

Documentation

Add basic documentation for public interfaces

td-idf

For easier use with document classification

Bernouli Naive Bayes

Another discrete model. see #3

Continuous Integration

change from struct to class, any issues?

Hi Fabien,
thanks for the great code.
For a personal project, I need to use BayesianClassifier through a computed var, with get access only. Unfortunately, because it's a struct, and it's mutated when calling .observe(), the compiler won't let me do it....

Changing struct to class seems to do the job. I'm however unsure about the consequence.

Would you have any recommendations?
thanks!
-M