This Python package is intended to provide a uniform schema for common machine learning applications, as well as a set of decorators that can be used to aid in web based ML prediction applications.
The input and output decorators offered by this package are design to simplify the process of prediction in a web based application. They each provide support for schema generation based on a provided sample input, with the idea being that this schema can then be taken and embedded into an API Swagger specification. Additionally, the input decorator provides support for type conversion at runtime from a JSON based input into the user specified sample type, to allow for easy conversion of input over the wire into the datatype that the decorated function expects. These decorators can be nested with each other and with other decorators as desired.
The package provides support for generating schema based on example input provided to the input and output decorators. This is intended to introduce a uniform conversion between a JSON format that can be embedded into a swagger specification and the in memory Python objects which may or may not have a built-in JSON representation. Currently it only supports OpenAPI 2.x
The input decorator provides support to convert input that is passed to the decorated function from a JSON type into the type specified by the provided sample. This allows the function to be called in a web based manner without needing to convert the data from over the wire either prior to the function call or as a part of the function handling. If the provided input is already of the sample type, the decorator is a no-op. Each currently supported type offers options on how much conversion and input enforcement should be done at runtime. Currently the output decorator does no form of type conversion.
Currently the package supports generation for Numpy, Pandas, and Spark types, as well as standard Python types. These
types are defined here. Custom types can be implemented by extending the
AbstractParameterType and overriding the
deserialize_input and input_to_swagger methods. It also supports nested dict or list inputs in case detailed
description inside or data type conversion is desired. The item inside dict or list will be treated as a valid parameter
if and only if they are of subtype of AbstractParameterType, which is an iterative definition.
Some sample usage for the decorators and each of the supported types can be found in the test resources.
This package is available for install via PyPi. The package supports dependency install via pip extras, so as to not bloat the user environment. Currently available extras are 'numpy-support', 'pandas-support', and 'spark-support'.
This project follows the Microsoft standard contribution guidelines. More information can be found here.
In order to work on this project, you will need a working Python 3 environment (virtualenv recommended). Package dependencies are specified in setup.py, as well as package extras. These extras are defined separately to allow users to install only the packages they want, to not bloat their environment.
Please refer to LICENSE for package licensing information.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [email protected] with any additional questions or comments.