Skip to content

real_osm_benchmark_result.md

Latest commit

 

History

History
41 lines (30 loc) · 2.37 KB

real_osm_benchmark_result.md

File metadata and controls

41 lines (30 loc) · 2.37 KB
Raw

Real OSM / GeoJSON Benchmark Result

This report is intentionally conservative. The committed result uses the local data/roads.geojson demo extract because no large real OSM city extract is committed to the repository. It is a toy/local sample and does not represent city-level performance.

To reproduce with a real extract:

python -m scripts.prepare_osm_extract --input-osm path/to/extract.osm --output datasets/osm_samples/prepared_roads.geojson
python -m benchmarks.spatial_index_benchmark --dataset osm_roads_sample --items 100000 --queries 1000 --indexes brute_force,grid,quadtree,rtree,str_rtree,kdtree --query-types bbox --output docs/assets/real_osm_spatial_index_benchmark.json

Dataset

  • Data source: data/roads.geojson via scripts.prepare_osm_extract --input-geojson
  • Road segments: 13
  • BBox: [116.39, 39.9, 116.41, 39.92]
  • Total length: 13,190.69 m
  • Query count: 100 bbox queries
  • Hardware environment: Windows 11 10.0.26100, Intel64 Family 6 Model 183 Stepping 1, 32 logical CPUs
  • Python: 3.12.7, Anaconda build, MSC v.1929 64-bit

Results

Index Build ms Memory bytes Mean candidates p50 ms p95 ms p99 ms Recall False positive Correctness
brute_force 0.004 1248 0.77 0.0012 0.0015 0.0022 1.000 0.000 pass
grid 0.060 1248 0.77 0.0017 0.0027 0.0041 1.000 0.000 pass
quadtree 0.036 1248 0.77 0.0020 0.0033 0.0541 1.000 0.000 pass
rtree 0.042 1248 0.77 0.0012 0.0018 0.0021 1.000 0.000 pass
str_rtree 0.023 1248 0.77 0.0012 0.0018 0.0020 1.000 0.000 pass
kdtree 0.027 1248 0.77 0.0011 0.0013 0.0014 1.000 0.000 pass

Correctness is measured against brute-force bbox query results for the same sampled queries. The benchmark JSON and generated table live in:

  • docs/assets/real_osm_extract_summary.json
  • docs/assets/real_osm_spatial_index_benchmark.json
  • docs/assets/real_osm_spatial_index_benchmark.md

Interpretation

The committed sample proves the benchmark path is executable and correctness-checked. It does not prove large-city performance. For resume or interview discussion, describe this as a city-scale-capable benchmark generator with a reproducible local report, then bring a separately generated real extract report if available.