Skip to content

Welcome!

This is a guide to Triage, a machine learning / data science tool initially developed at the Center for Data Science and Public Policy (DSaPP) at the University of Chicago and now being maintained at Carnegie Mellon University.

Triage helps build ML systems for two common problems: (a) Early warning systems (EWS or EIS), (b) resource prioritization (a.k.a "an inspections problem"). These problems require careful thought and design and their formulation and implementation are often done incorrectly.

Info

This tutorial is in sync with the latest version of triage. At this moment v4.2.0.

How you can help to improve this tutorial

If you want to contribute, please follow the suggestions in the triage’s github repository.

Why Dirty Duck??#

There is a famous (and delicious) peking duck restaurant in Chicago called Sun Wah. We love that place, and as every restaurant in Chicago area, it gets inspected, so the naming is an homage to them.

Who is this tutorial for?#

We created this tutorial with two roles in mind:

  • Data scientists/ML practitioners who want to focus on the problem they are tackling, and not on the nitty-gritty details about how to configure and setup a Machine learning pipeline, model governance, reproducibility, model selection, etc.

  • analytical policy team without too deep of a technical/engineering background who want to learn how to formulate their policy problems as Machine Learning problems.

How to use this tutorial#

First, clone this repository on your machine

git clone https://github.com/dssg/triage

Second, in the cloned repository's top-level directory run

./tutorial.sh up

This will take several minutes the first time you do it.

After this, you may decide to do the quickstart tutorial.

Before you start#

What you need for this tutorial#

Install Docker CE and Docker Compose. That's it! Follow the links for the installation instructions.

Note that if you are using GNU/Linux you should add your user to the docker group following the instructions at this link.

At the moment only operating systems with *nix-type command lines are supported, such as GNU/Linux and MacOS. Recent versions of Windows may also work.