PyObjC CoreML Framework

Warnings

• breaking PyObjC 12.0 dropped support for Python 3.9. Users must ensure their environment uses Python 3.10 or later. PyObjC 11.0 also dropped support for Python 3.8.
  Fix: Upgrade Python to version 3.10 or higher.
• breaking PyObjC 11.1 aligned its memory management behavior for `init` methods with `clang`'s Automatic Reference Counting (ARC). Methods in the `init` family now correctly steal a reference to `self` and return a new one. This change might affect code relying on previous PyObjC reference counting behavior for initializers.
  Fix: Review and update code interacting with Objective-C `init` methods to ensure correct reference handling, especially for complex object graphs or custom subclasses.
• gotcha In PyObjC 10.3, a change broke the ability to use Python's `__init__` when a user also implements `__new__` for an Objective-C subclass. While 10.3.1 partially re-introduced this, classes relying on `__new__` provided by PyObjC still cannot use `__init__`.
  Fix: If subclassing Objective-C classes, avoid implementing both `__new__` and `__init__` in Python for versions 10.3 and later, or carefully test behavior when a user-defined `__new__` is present.
• breaking The `IMServicePlugIn` bindings were completely removed in PyObjC 10.0. This framework was deprecated by Apple in macOS 10.13 and removed in macOS 14.
  Fix: Migrate away from the `IMServicePlugIn` framework. There is no direct replacement in newer macOS versions via PyObjC.
• gotcha While PyObjC 11.0 introduced *experimental* support for free-threading (`PEP 703`) with Python 3.13, PyObjC 10.3 explicitly stated it does *not* support experimental free-threading in Python 3.13. Users should exercise caution and thoroughly test when using free-threading with PyObjC, especially with specific framework bindings.
  Fix: Avoid relying on Python 3.13's experimental free-threading features with PyObjC unless explicitly confirmed as stable and supported for your specific PyObjC version and frameworks.

Install

• pip install pyobjc-framework-coreml Install latest version

Imports

• CoreML

  import CoreML

  All CoreML classes and functions are typically exposed under the `CoreML` top-level module.
• MLModel

  from CoreML import MLModel

  Specific CoreML classes can be imported directly for brevity.

Quickstart

This quickstart demonstrates how to import the `CoreML` framework and attempt to load an `MLModel` using PyObjC. It highlights the typical PyObjC pattern for handling Objective-C methods that return errors by reference. A functional example would require an actual `.mlmodel` file.

import CoreML
import Foundation # Often needed for NSURL, NSBundle etc.
import os

# --- This example demonstrates attempting to load a CoreML model ---
# In a real scenario, 'your_model.mlmodel' would be a path to an actual Core ML model file.
# For demonstration, we'll try a dummy path or look for a model in the app bundle.
# Replace with the actual path to your .mlmodel file or ensure it's in your app bundle.

# Option 1: Direct path (replace with actual path)
model_path = os.environ.get('COREML_MODEL_PATH', '/tmp/your_model.mlmodel')

# Option 2: Attempt to find in the main bundle (common for macOS apps)
# mainBundle = Foundation.NSBundle.mainBundle()
# model_path_from_bundle = mainBundle.pathForResource_ofType_("YourModelName", "mlmodel")
# if model_path_from_bundle:
#     model_path = model_path_from_bundle

print(f"Attempting to load CoreML model from: {model_path}")

try:
    # CoreML methods often follow Objective-C conventions, e.g., 'methodName_error_'
    # for methods that take an NSError** parameter in Objective-C.
    # The PyObjC bridge translates this into a tuple (result, error_object).
    model, error = CoreML.MLModel.modelWithContentsOfURL_error_( 
        Foundation.NSURL.fileURLWithPath_(model_path), None
    )
    
    if error:
        print(f"Error loading model: {error.localizedDescription()}")
    elif model:
        print(f"Successfully loaded CoreML model: {model}")
        print(f"Model description: {model.modelDescription()}")
        # You can now interact with the model, e.g., for predictions:
        # input_features = CoreML.MLFeatureProvider.alloc().init()
        # prediction, pred_error = model.predictionFromFeatures_error_(input_features, None)
        # if not pred_error: print(f"Prediction: {prediction}")
    else:
        print("Failed to load model without explicit error or model object.")

except Exception as e:
    print(f"An unexpected Python exception occurred: {e}")

print(f"\nAccessing MLModel class directly: {CoreML.MLModel}")