AW: ML Model Serialization Utilities

A Python library for managing, serializing, and deserializing machine learning models and their attributes.

• Extract model attributes and parameters into serializable dictionaries
• Convert ML models to JSON-friendly formats
• Save and load model parameters to/from JSON
• Handle NumPy arrays, SciPy sparse matrices and other complex data types
• Create transformers from non-transformer objects

• trailing_underscore_attributes_with_include_and_exclude(obj, include=(), exclude=()): Get attributes with trailing underscore, with custom inclusion/exclusion
• get_model_attributes(model, include=(), exclude=(), model_name_as_dict_root=True, as_is_types=default_as_is_types): Export model parameters to a dict

• get_model_attributes_dict_for_json(model, include=(), exclude=(), model_name_as_dict_root=True, as_is_types=default_as_is_types): Get model attributes as a JSON-compatible dictionary
• export_model_params_to_json(model, include=(), exclude=(), model_name_as_dict_root=True, as_is_types=default_as_is_types, filepath='', version=None, include_date=False, indent=None): Export model parameters to JSON file or string
• import_model_from_spec(spec, objects={}.copy(), type_conversions=(), field_conversions={}.copy(), force_dict_wrap=False): Reconstruct a model from specification dictionary

• json_friendly_dict(obj): Convert Python objects to JSON-serializable format
• NumpyAwareJSONEncoder: JSON encoder that handles NumPy arrays and other special types

• ExtrapolateTransformation(transformer, extrapolator=LinearRegression()): Wrap a transformer to provide transform method using a regression model

from sklearn.cluster import KMeans
import numpy as np
from aw import export_model_params_to_json, import_model_from_spec
import json

# Create a model
kmeans = KMeans(n_clusters=3)
kmeans.fit(np.random.rand(100, 5))

# Export model parameters to JSON
json_str = export_model_params_to_json(kmeans)

# Load the JSON string
model_spec = json.loads(json_str)

# Recreate the model from spec
reconstructed_model = import_model_from_spec(model_spec, objects={'KMeans': KMeans})

• NumPy
• SciPy
• scikit-learn
• dill

[License information goes here]