CoreML + CreateML Kit

Complete machine learning toolkit for iOS 18+ / macOS 15+ with CoreML inference and CreateML training.

🤖 Features

CoreML Inference

• Image classification
• Text classification
• Batch processing
• Actor-based concurrency

CreateML Training

• Image classifier training
• Text classifier training
• Tabular data training
• Data augmentation

Model Evaluation

• Accuracy metrics
• Confusion matrix
• Performance testing

📦 Installation

dependencies: [
    .package(url: "https://github.com/durellwilson/coreml-createml-kit.git", from: "1.0.0")
]

🚀 Quick Start

Image Classification

import CoreMLCreateMLKit

let classifier = try ImageClassifier(modelName: "MyImageClassifier")

let image = UIImage(named: "test")!
let results = try await classifier.classify(image)

for result in results {
    print("\(result.label): \(result.confidence)")
}

Text Classification

let classifier = try TextClassifier(modelName: "SentimentClassifier")

let prediction = await classifier.classify("This is amazing!")
print("Sentiment: \(prediction?.label ?? "unknown")")

// Batch processing
let texts = ["Great!", "Terrible", "Okay"]
let predictions = await classifier.batchClassify(texts)

Train Image Classifier (macOS)

#if os(macOS)
import CreateML

let trainer = CreateMLTrainer()

let model = try await trainer.trainImageClassifier(
    trainingData: URL(fileURLWithPath: "/path/to/training"),
    validationData: URL(fileURLWithPath: "/path/to/validation"),
    iterations: 25
)

try model.write(to: URL(fileURLWithPath: "/path/to/output.mlmodel"))
#endif

Train Text Classifier

let trainer = CreateMLTrainer()

let model = try await trainer.trainTextClassifier(
    trainingData: URL(fileURLWithPath: "/path/to/data.csv"),
    textColumn: "text",
    labelColumn: "sentiment"
)

try model.write(to: URL(fileURLWithPath: "/path/to/sentiment.mlmodel"))

Evaluate Model

let evaluator = ModelEvaluator()

let predictions: [(predicted: String, actual: String)] = [
    ("cat", "cat"),
    ("dog", "dog"),
    ("cat", "dog")
]

let metrics = await evaluator.evaluate(predictions: predictions)
print("Accuracy: \(metrics.percentage)")

let matrix = await evaluator.confusionMatrix(predictions: predictions)
print("Confusion Matrix: \(matrix)")

🏗️ Project Structure

Training Pipeline (macOS)

// 1. Prepare data
let trainingURL = URL(fileURLWithPath: "/path/to/training")

// 2. Train model
let trainer = CreateMLTrainer()
let model = try await trainer.trainImageClassifier(
    trainingData: trainingURL,
    iterations: 50
)

// 3. Evaluate
let testData = loadTestData()
let predictions = try await classify(testData, with: model)
let metrics = await evaluator.evaluate(predictions: predictions)

// 4. Export
try model.write(to: outputURL)

Inference Pipeline (iOS/macOS)

// 1. Load model
let classifier = try ImageClassifier(modelName: "MyModel")

// 2. Classify
let results = try await classifier.classify(image)

// 3. Process results
let topResult = results.first!
if topResult.confidence > 0.8 {
    print("High confidence: \(topResult.label)")
}

🎯 Use Cases

Image Recognition App

struct ContentView: View {
    @State private var results: [Classification] = []
    let classifier = try! ImageClassifier(modelName: "MobileNetV2")
    
    var body: some View {
        VStack {
            Image("photo")
            
            Button("Classify") {
                Task {
                    results = try await classifier.classify(image)
                }
            }
            
            List(results) { result in
                HStack {
                    Text(result.label)
                    Spacer()
                    Text("\(Int(result.confidence * 100))%")
                }
            }
        }
    }
}

Sentiment Analysis

let classifier = try TextClassifier(modelName: "SentimentModel")

func analyzeFeedback(_ feedback: [String]) async -> [String: Int] {
    let predictions = await classifier.batchClassify(feedback)
    
    var counts: [String: Int] = [:]
    for prediction in predictions {
        counts[prediction.label, default: 0] += 1
    }
    return counts
}

🧪 Testing

func testClassifier() async throws {
    let classifier = try ImageClassifier(modelName: "TestModel")
    let testImage = UIImage(named: "test")!
    
    let results = try await classifier.classify(testImage)
    
    XCTAssertFalse(results.isEmpty)
    XCTAssertGreaterThan(results[0].confidence, 0.5)
}

📊 Performance

• Image classification: ~50ms on device
• Text classification: ~10ms
• Batch processing: Parallel execution
• Memory efficient: Actor isolation

🤝 Contributing

Part of Detroit's ML education initiative!

📝 License

MIT License

---

Production ML with CoreML + CreateML 🤖