racket 0.3.9

Overview

Let’s face it. Building models is already challenging enough. But putting them into production is usually a big enough challenge to grant the employment of an entire separate team. The goal of the project is removing (or at least softening) the dependency on machine learning engineers and devops, enabling data scientist to go from concept to production in minutes.

Presented at PyData: video, slides

Features

• Easy integration with TensorFlow Serving and Keras

• RESTful interface with interactive Swagger documentation

• Model introspection: ability to view model performance and input requirements

• Ability to deploy automatically different models with a single command

• Rich CLI capabilities, going from project scaffolding to training, serving, and dashboarding

• Small codebase, statically typed with mypy, and extensive docstrings

• Coming Soon ^TM: Web-ui for managing, introspecting, and deploying models.

Demo

Roadmap

• Web dashboard for model management and introspection

• Support for Pytorch using ONNX

• Path to production: docker-based deployments to major cloud providers

• Security capabilities with SSL encryption

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.

The icon was created by smashicons.

0.1.0 (2018-11-02)

• First release on PyPI.

0.2.0 (2018-11-13)

• Major feature implementation and documentation

• Static typing

• Testing - 78% coverage

0.3.0 (2018-11-20)

• Major internals refactoring

• API unchanged, although external API was made more clear and documented