README.md

metadata

license: cc-by-4.0
language:
  - en
  - fr
tags:
  - smart-building
  - benchmark
  - timeseries
  - graph-database
  - postgresql
  - memgraph
  - sparql
  - building-automation
task_categories:
  - time-series-forecasting
  - graph-ml
size_categories:
  - 100M<n<1B
configs:
  - config_name: default
    data_files:
      - split: graph_nodes
        path: data/graph/nodes.parquet
      - split: graph_edges
        path: data/graph/edges.parquet
      - split: timeseries
        path: data/timeseries/*.parquet

BaseType Benchmark Dataset

Reference dataset for benchmarking database paradigms on smart building data.

Paper

TBD

Quick Start

from basetype_benchmark.dataset.huggingface import load_benchmark_data

# Load a specific profile
data = load_benchmark_data(scale="medium", duration="1m")

print(f"Nodes: {len(data['nodes'])}")
print(f"Edges: {len(data['edges'])}")
print(f"Timeseries points: {len(data['timeseries'])}")

Available Profiles

Scale (Taille du graphe)

Scale	Description	Points	Buildings
small	Single building	~50k	1
medium	Small campus	~100k	5
large	Full campus	~500k	10+

Duration (Priode temporelle)

Duration	Description	Days
1d	One day	1
1w	One week	7
1m	One month	30
6m	Six months	180
1y	One year	365

Dataset Structure

data/
 graph/
    nodes.parquet     # ~50k nodes across 9 domains
    edges.parquet     # ~200k relationships
 timeseries/
    2024-01.parquet   # Partitioned by month
    2024-02.parquet
    ...

Domains Covered

1. Spatial: Site, Building, Floor, Space, Zone
2. Equipment: Systems, Equipment, Sensors, Actuators
3. Energy: Meters (tree distribution), EnergyZones
4. IT/Datacenter: Servers, Racks, Network devices
5. Audiovisual: AV Systems, Displays, Projectors
6. Parking: Zones, Spots, Charging stations
7. Security: Access points, Cameras, Alarms
8. Organization: Departments, Teams, Persons
9. Contractual: Contracts, Providers, Work orders

Schema

nodes.parquet

Column	Type	Description
node_id	string	Unique identifier
node_type	string	Type (Building, Equipment, etc.)
name	string	Human-readable name
building_id	int	Building ID for scale filtering (0=cross-building)
properties	json	Domain-specific attributes

edges.parquet

Column	Type	Description
src	string	Source node ID
dst	string	Target node ID
rel	string	Relationship type

timeseries/*.parquet

Column	Type	Description
point_id	string	Reference to Point node
timestamp	timestamp	Measurement time
value	float64	Measured value
building_id	int	Building ID for scale filtering
year_month	string	Partition key (YYYY-MM)

Scale Filtering

The dataset uses building_id for scale-based extraction:

• small: building_id = 1 (single building)
• medium: building_id IN (1,2,3,4,5) (5 buildings)
• large: all buildings (no filter)
• building_id = 0: cross-building entities (Site, Organization, etc.)

Benchmark Usage

# Run benchmark with specific profile
python -m basetype_benchmark.run --scale=medium --duration=1m

# Results include:
# - Query latencies (p50, p95, min, max)
# - Memory consumption (steady state, peak)
# - CPU usage (average, spikes)
# - Energy estimation (via RAPL/CPU-time)

Author

Antoine Debienne

Citation

@dataset{basetype_benchmark_2025,
  author = {Antoine Debienne},
  title = {BaseType Benchmark Dataset},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/synaptikAD/basetype-benchmark}
}

License

CC-BY-4.0