Skip to content

v3.3.0 — Core ML Pipelines, .mlpackage, VectorStore, DataFrame.toSQL & Local Embeddings

Choose a tag to compare

View all tags
@Nodibell Nodibell released this 21 Aug 20:23
· 16 commits to main since this release
v3.3.0
d553df3

🚀 SwiftSci 3.3.0 — Release Notes

SwiftSci v3.3.0 expands Apple Silicon scientific computing and on-device MLOps with composite Core ML pipeline export, modern .mlpackage directory bundles, an in-memory vector database (VectorStore), high-speed tabular database ingestion (DataFrame.toSQL), and offline dense text embeddings.

🌟 Key Deliverables & Highlights

1. 🍏 Composite Core ML Pipeline Serializer (SwiftML)

  • PipelineClassifier & PipelineRegressor: Implemented native binary Protobuf export for multi-stage models (fields 200/201 in Apple's Model.proto specification) via CoreMLExporter.exportBinaryPipelineClassifier and CoreMLExporter.exportBinaryPipelineRegressor.
  • Enables chaining preprocessors (StandardScaler, OneHotEncoder) with classifiers (RandomForestClassifier, MLPClassifier, LogisticRegression) into a single Core ML artifact.

2. 📦 Modern .mlpackage Directory Bundle Exporter (SwiftML)

  • writeMLPackage: Generates standard .mlpackage directory bundles containing valid Manifest.json and nested Data/com.apple.CoreML/model.mlmodel payloads.
  • Conformed CoreMLExportable with a default writeMLPackage(to:author:description:) method for all SwiftML models.

3. 🧠 In-Memory VectorStore Index (SwiftCluster)

  • Accelerate-Optimized Vector Database: High-throughput in-memory vector storage (VectorStore) with SIMD-vectorized distance and similarity metrics:
    • .cosineSimilarity (vDSP_dotprD + vDSP_svesqD)
    • .dotProduct
    • .euclideanDistance (vDSP_distancesqD)
  • Thread-safe (@unchecked Sendable with NSLock) with Top-K search, batch insertions (addBatch), and item inspection.

4. 🗄️ Batch Database Ingestion & TLS Security (SwiftDatabase)

  • DataFrame.toSQL: High-speed tabular bulk insertion into SQLite, PostgreSQL, and MySQL tables with .append, .replace, and .failIfExists modes and automatic schema type inference.
  • SSLMode: Configurable TLS/SSL encryption (.disable, .prefer, .require) with query string parsing (?sslmode=require, ?ssl=true) for remote PostgreSQL and MySQL instances.

5. 🔤 Local Dense Text Embedding Engine (SwiftNLP)

  • LocalEmbeddingEngine: Fast, pure-Swift dense text embedding generator (128-D/256-D L2-normalized vector embeddings, $|v|_2 = 1.0$) operating 100% locally and offline on Apple Silicon, directly interoperable with SwiftCluster.VectorStore.

6. 📚 100% DocC Documentation & Quality Guardrails

  • 100.00% Public API Coverage: 1,356 of 1,356 public symbols documented across all 14 modules.
  • CI Test Suite: 380+ unit tests passing with zero regressions.

📦 Installation (Swift Package Manager)

dependencies: [
    .package(url: "https://github.com/Nodibell/SwiftSci.git", from: "3.3.0")
]
Assets 2
Loading