Usage¶
One month, in a few lines¶
1 2 3 4 5 | |
run_pipeline downloads that month's file from SLMPD, cleans it, and keeps only records that fall inside the City of St. Louis. It returns a CrimeDataPackage with three parts:
| Part | Meaning |
|---|---|
pkg.df |
The cleaned table, one row per offense |
pkg.month, pkg.year |
The month and year it covers |
If SLMPD has not published the month yet (or the download fails) you get an empty table and a warning, not an exception.
Months from January 2024 onward are supported. January to April 2024 were published under one-off file names; the package knows their addresses (see SPECIAL_URLS in config.py).
What the cleaned table contains¶
| Column | Meaning |
|---|---|
inc_# |
Incident number |
date_time |
When the offense occurred (see the note on missing times in the FAQ) |
nibrs_cat, nibrs_code |
NIBRS category and code |
offense, inc_desc |
Offense and incident description |
off_type |
Crime against Person, Property or Society |
firearm |
Whether a firearm was used |
address, nbhd, nbhd_num, district, ward |
Where it happened (ward is assigned from the coordinates) |
lat, lon |
Coordinates |
Saving a month¶
1 2 3 | |
Files are named like stl_crime_data_04_2026.csv. Each function returns True on success and False (with a logged error) otherwise. The output folder must already exist.
Building the full history (from a clone)¶
1 2 | |
run_pipeline.py always takes the previous calendar month. For a specific month, call it from Python: from run_pipeline import run_monthly_etl; run_monthly_etl(month=9, year=2026).
These write the master dataset to data/processed/slmpd_crime_data.parquet (GeoParquet) and a GeoJSON copy for GIS users. Rebuild only the Explorer's data feed from the master with:
1 | |
See data/README.md in the repository for the folders and what is committed.
Notebooks¶
Two example notebooks are on the Examples menu: a step-by-step walkthrough of the pipeline, and a hotspot demo.