serialzy's Introduction

serialzy

Serialzy is a library for python objects serialization into portable and interoperable data formats (if possible).

Example

Suppose you have a catboost model:

from catboost import CatBoostClassifier

model = CatBoostClassifier()
model.fit(...)

Firstly you should find a proper serializer for the catboost model type or the corresponding data format:

from serialzy.registry import DefaultSerializerRegistry

registry = DefaultSerializerRegistry()
serializer = registry.find_serializer_by_type(type(model)) # registry.find_serializer_by_data_format("cbm")

Serializers have several properties:

serializer.available()      # can be used in the current environment
serializer.requirements()   # libraries needed to be installed to use this serializer
serializer.stable()         # has portable data format

Serializers can provide data format and schema for a type:

serializer.data_format()
serializer.schema(type(model))

Serialization:

with open('model.cbm', 'wb') as file:
    serializer.serialize(model, file)

Deserialization:

with open('result', 'rb') as file:
    deserialized_obj = serializer.deserialize(file)

List of supported libraries for stable serialization:

Library	Types	Data format
Python std lib	int, str, float, bool, None	string representation
Python std lib	List, Tuple	custom format
CatBoost	CatBoostRegressor, CatBoostClassifier, CatBoostRanker	cbm
CatBoost	Pool	quantized pool
Tensorflow.Keras	Sequential, Model with subclasses	tf_keras
Tensorflow	Checkpoint, Module with subclasses	tf_pure
LightGBM	LGBMClassifier, LGBMRegressor, LGBMRanker	lgbm
XGBoost	XGBClassifier, XGBRegressor, XGBRanker	xgb
Torch	Module with subclasses	pt
ONNX	ModelProto	onnx

Recommend Projects