hollance / sefr-swift

The SEFR classifier implemented in Swift

SEFR in Swift

This is the source code that accompanies my blog post The SEFR classifier.

SEFR is a binary classifier. To use it on a binary classfication task:

let sefr = SEFR()
sefr.fit(examples: X_train, targets: y_train)
let y_pred = sefr.predict(examples: X_test)

You can use it for multiclass tasks by using a simple wrapper that performs one-vs-rest:

let model = SEFRMulticlass()
model.fit(examples: X_train, targets: y_train)
let y_pred = model.predict(examples: X_test)

The demo program is a simple Swift script that runs on macOS. It uses the Vision framework to generate "feature prints" for a set of training and test images. Each feature print consists of 2048 numbers. It trains SEFR on these feature prints.

To run the demo program, open a Terminal on macOS and type:

$ cat SEFR.swift Demo.swift | swift -

Credits

The images used in the dataset were hand-picked from Google Open Images Dataset V6. For full credits and license terms, see here.

The source code is licensed as MIT.

My implementation is based on the following sources:

• SEFR: A Fast Linear-Time Classifier for Ultra-Low Power Devices
• Original Python implementation
• Multiclass version in Python