Swift Not Good For Training Machine Learning Models
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
After working extensively with Create ML, the author concludes that Swift is well-suited for deploying ML models but not for training them. Key pain points include verbose data preprocessing compared to Python's pandas/scikit-learn, Playground instability with linker errors, and a less mature ecosystem overall. The recommended workflow is to train models in Python, convert them to Core ML format using coremltools, then deploy on-device or via a Swift server backend (Vapor/Hummingbird). Code examples show the conversion process and a Vapor route serving car price predictions from a Core ML model. One caveat: hosting Core ML on a server requires macOS, not Linux.
