DannHiroaki/CMAB-Spatial-Join-0.08B

Introduction

A lightweight axis-aligned rectangle (AABB) benchmark derived from the CMAB building rooftop dataset, designed for evaluating spatial join pipelines (indexing, vectorized filtering, output materialization).

This dataset stores base AABBs and expanded (influence) AABBs for each building under 4 workload levels. It is organized as Parquet shards, partitioned by level and province.

Dataset construction details and the reference builder are available at: https://github.com/DANNHIROAKI/CMAB-Spatial-Join-0.08B-Builder

How to Use

1) Download a small slice (recommended)

from huggingface_hub import snapshot_download
# Download only what you need (e.g., Level-1 + Beijing) with snapshot_download
local_dir = snapshot_download(
    repo_id="DannHiroaki/CMAB-Spatial-Join-0.08B",
    repo_type="dataset",
    allow_patterns=[
        "dataset_metadata.json",
        "file_manifest.parquet",
        "summary_stats.parquet",
        "level_1/province=beijing/*.parquet",
    ],
)
print("Downloaded to:", local_dir)

2) Read a shard with Polars

import polars as pl
from pathlib import Path

p = Path(local_dir) / "level_1" / "province=beijing" / "data_00001.parquet"
df = pl.read_parquet(p)

print(df.select(["building_uid","func","level","d_m","xmin","ymin","xmax","ymax","exmin","eymin","exmax","eymax"]).head())

3) Query with DuckDB (fast ad-hoc analytics)

import duckdb
from pathlib import Path

glob_path = str(Path(local_dir) / "level_1" / "province=beijing" / "*.parquet")

con = duckdb.connect()
n = con.execute(f"""
  SELECT COUNT(*) 
  FROM read_parquet('{glob_path}')
  WHERE func = 'Residential'
""").fetchone()[0]

print("Residential rows (level_1, beijing):", n)

4) Use file_manifest.parquet to discover shards

import polars as pl
from pathlib import Path

mf = pl.read_parquet(Path(local_dir) / "file_manifest.parquet")
print(mf.select(["level","province","path","num_rows"]).sort(["level","province"]).head(20))

Attribution

This benchmark is derived from CMAB:

• Paper: CMAB: A Multi-Attribute Building Dataset of China

  @article{Zhang2025SciData,
    author  = {Zhang, Y. and Zhao, H. and Long, Y.},
    title   = {{CMAB: A Multi-Attribute Building Dataset of China}},
    journal = {Scientific Data},
    volume  = {12},
    number  = {430},
    year    = {2025},
    doi     = {10.1038/s41597-025-04730-5},
    url     = {https://doi.org/10.1038/s41597-025-04730-5}
  }
  

• Dataset: CMAB-The World’s First National-Scale Multi-Attribute Building Dataset

  @misc{Zhang2025CMAB,
    author       = {Zhang, Yecheng and Zhao, Huimin and Long, Ying},
    title        = {{CMAB-The World's First National-Scale Multi-Attribute Building Dataset}},
    year         = {2025},
    month        = apr,
    publisher    = {figshare},
    doi          = {10.6084/m9.figshare.27992417},
    url          = {https://doi.org/10.6084/m9.figshare.27992417},
    howpublished = {dataset}
  }
  

If you use this dataset in research, please cite the CMAB paper/dataset and reference this benchmark repository.