Skip to content

Latest commit

 

History

History
50 lines (31 loc) · 2.42 KB

README.md

File metadata and controls

50 lines (31 loc) · 2.42 KB
Raw

NBA Player Analytics

This analytics pipeline is composed of the following steps:

  • EXTRACT - Data is procured from the NBA's API (see src/api_helpers.py)
  • LOAD - Data is loaded into BigQuery via a custom wrapper (see src/utilities/bigquery.py)
  • TRANSFORM - Raw BigQuery data is cleaned and reshaped with dbt (see nba_dbt/)

Source Code

You'll need a Google service account stored locally - ./service_accounts/nba-player-analytics-service.json ... this is used by every service!

Infrastructure

The core cloud infrastructure is managed via Terraform in the infrastructure/ directory. Configuration, production and development datasets, a production Cloud Run service, and all the relevant IAM permissions are managed here. To run this locally, ensure you have a infrastructure/local.json file defined like:

{
    "USER_EMAIL": "<FILL>",
    "SERVICE_ACCOUNT": "<FILL>",
    "DOCKER_IMAGE_NAME": "<FILL>"
}

Deployment

You'll find production and development Dockerfiles in the deploy/ directory. You can docker compose up --build -d from the root of this project to build the dev environment; I've included a shell script to open an interactive terminal to test things out.

The big difference here is that DEV sleeps infinitely, so it just sits idly and waits for a command. PROD runs a series of commands (defined in deploy/entrypoint.sh) to authenticate the shell and connect it to Prefect Cloud.

Extract + Load

This is all wrapped in src/run.py. There are optional CLI arguments to run this locally (outside the cloud context), with a debugger, fully refreshing the data, etc... see src/utilities/cli.py for a full list.

This project IS hardcoded to run Knicks data, because frankly that's all I cared to do!

Transform

All of the data transformations are handled by dbt - see nba_dbt/README.md for a quick mapping. Transformations happen in staging/, aggregations and splits happen in clean/ and individual players are modeled in player_summaries/

Fun Next Steps

  • Pull this data into a React app to serve as a dashboard. Player dropdown, stat dropdown, all team stats, etc.
  • Train some simple ML models to predict team success as a factor of individual stats and conditions
    • Team offensive rating when Jalen plays > 30 minutes?
    • Defense against top-4 seeds when playing back to back games?