Aug 13: Context layer webinar
Chalk home page
  1. Feature Engine
  2. Quick Start

This page starts where the Setup Guide ends: you have a project that chalk init created, and the CLI is logged in. It walks through connecting your data sources, defining feature classes and resolvers, and deploying and querying them.

​
Configure data sources

Your resolvers load data from the data sources that you add in the Chalk dashboard.

  1. Open the dashboard with chalk dashboard, or go to https://chalk.ai/projects. The dashboard shows every environment that your login has access to.
  2. In the sidebar, select Data Sources and add each data source that you will work with. After you save a data source, select Test Data Source in the upper-right corner of its configuration view to check the connection.
  3. Create a src/datasources.py file that refers to the data sources you added: 
    root_directory/
├── src/
│  └── datasources.py
├── chalk.yaml
├── README.md
└── pyproject.toml

For example, for a PostgreSQL data source named PG, datasources.py contains:

from chalk.sql import PostgreSQLSource

pg = PostgreSQLSource(name='PG')

For the settings of each kind of data source, see Integrations.

​
Define feature classes and resolvers

A feature class is a Python class of features, and a resolver tells Chalk how to compute feature values. Each feature that you define is the output of a resolver.

We recommend starting with a small feature class and adding to it, testing as you go. A project with several feature classes, resolvers, and tests looks something like this:

root_directory/
├── src/
│  ├── resolvers/
│  │  ├── .../
│  │  ├── __init__.py
│  │  └── pipelines.py
│  ├── __init__.py
│  ├── datasources.py
│  └── feature_sets.py
├── tests/
│  └── ...
├── .chalkignore
├── chalk.yaml
├── README.md
└── pyproject.toml

For the kinds of features and resolvers that you can write, see Features and Resolvers. For help structuring your feature classes and resolvers, ask in your support channel.

​
Deploy and query

Deploy your features and resolvers with chalk apply. During development, deploy to a branch with chalk apply --branch <branch-name>, so that several people can work in one environment at the same time. Deploying without --branch requires the deploy.create permission, which the built-in Admin and Owner roles include. For the permissions each task needs, see Permissions.

After you deploy, query your features from the command line with chalk query, or from code with a Chalk client for Python, Go, Java, or TypeScript.

Repeat these steps (write, deploy, and query) until the feature values that you get back are the values you expect.

​
Orchestration

Once your features return the values you expect, decide how the rest of your machine learning platform uses them. You can trigger resolvers and run queries from other orchestrated pipelines, such as Airflow. You can also run resolvers on a cron schedule, and set staleness values on features so that the data you query meets your freshness requirements.

​
Further resources

For a tutorial that builds a fraud detection pipeline with Chalk, see Tutorial: Fraud Detection Pipeline.