Skip to content

Latest commit

 

History

History
63 lines (44 loc) · 1.97 KB

File metadata and controls

63 lines (44 loc) · 1.97 KB

Tutorial

This tutorial will walk you through the process of starting Clipper, creating an application, querying the application, and deploying models to improve application accuracy.

The tutorial uses the Clipper client management Python library for controlling Clipper. While this is the simplest way to manage Clipper, the client library is simply a convenience wrapper around a standardized management REST interface which can be queried with any REST client.

Setup

The tutorial runs Clipper in Docker containers and orchestrates them with Docker-Compose, so you must have Docker and Docker-Compose installed. The tutorial can be run in two modes, local or remote.

Running in local mode will start Clipper locally on your laptop. Running in remote mode will start Clipper on the machine you specify. When running the tutorial remotely, you must have SSH access to the machine. If running on EC2, you can use AMI ami-3ba0f05b, an Ubuntu image that has Docker and Docker-Compose installed.

The tutorial uses Python and Jupyter notebooks to interact with Clipper and train models. You can install the Python dependencies needed to run the tutorial with:

pip install -r requirements.txt

Finally, the tutorial uses the redis-cli tool to query Redis directly for some Clipper inspection commands. This tool is usually part of the Redis package.

On a Mac:

brew install redis

On Ubuntu 16.04:

apt-get install redis-tools

Note the Python and Redis-CLI dependencies must be installed for both the local and remote versions of the tutorial.

Running the tutorial

To start the tutorial, start the Jupyter server in this directory (examples/tutorial)

$ jupyter notebook

and open the tutorial_part_one.ipynb notebook. The tutorial is self-guided from there.

Happy serving!

If you run into any problems or find anything unclear while doing the tutorial, please file an issue or contact us (clipper-dev@googlegroups.com) so we can improve it.