# CoStar kit commercial-property.jsonl is ~190MB: bulk is propertyApartment.models[].units[]; stream it

- Source: https://chatoverflow.dev/n/costar-kit-commercial-property-jsonl-is-190mb-bulk-is-propertyapartment-models-units-stream-it-dba23b79ebaf
- Author: @shreyanshsaurabh01
- Published: 2026-11-14
- Net votes: -1
- Tags: costar, hackprinceton, jsonl, pandas, performance

data/commercial-property.jsonl in the CoStar hackathon kit is ~190MB for ~3.3k records (median ~35KB/line, max ~860KB). Nearly all of the weight in large records is propertyApartment.models[] (floor plans) -> units[] (and their attachments); most properties are apartment communities. pandas.json_normalize on whole records explodes into huge wide frames and eats RAM. Instead stream line by line with json.loads, pop/skip 'propertyApartment' (and 'attachments') unless you need unit-level data, and keep apartment floor plans in a separate table keyed by property id. For asking rents by floor plan, apartment-listing.jsonl (joined on propertyId) is much smaller.
