Request for bounded Hanoi demand data and dataset provenance
Dear DiTEC-WDN maintainers,
I am using the official DiTEC-WDN Hanoi benchmark for a research project on a water-distribution digital twin. I would appreciate your help clarifying the dataset's safe data-access options, node mapping, provenance, and engineering units.
My project uses simulated benchmark data, not real operational sensor measurements.
I have pinned the source to dataset revision 9a30ccd6309c03f524c0a4b5f5e6eb10b8b577c9, object hanoi_8GB_1Y/node_demand-0-dynamic_output.parquet.
For Phase 3, I need scenario 0 (tracer 60), node 2, with historical context at time IDs 6108–6131 and calibration targets at 6132–7445. Please exclude time IDs 7446–8759.
Most importantly, can you provide a source-faithful bounded export or a documented extraction process with evidence that the protected values at time IDs 7446 and later are not decoded or emitted? The current storage metadata does not establish that guarantee.
Could you also help clarify:
- The authoritative mapping from demand columns 2–32 to Hanoi junction IDs.
- The exact generator revision and scenario-generation provenance for this published object.
- The matching scenario-specific hydraulic input file and its revision or hash.
- The release-specific engineering units and meaning of
node_demand. - The source and output hashes, extraction procedure, and physical storage evidence for any bounded artifact.
Phase 4 multi-node data requirements are still under investigation, so I am not specifying a Phase 4 time range in this request.
If any information or safe export is unavailable, please let me know rather than infer it from a different release.
Thank you for your help.
Best regards,
sadiya
Hi Sadiya,
Thank you for your interest in our dataset. Please see the answers below.
A. I need scenario 0 (tracer 60), node 2, with historical context at time IDs 6108–6131 and calibration targets at 6132–7445. Please exclude time IDs 7446–8759.
You can simply download them using the HF interface load_dataset. I help you structure the code as an example, but please always check the filtered data:
# load the lib
from datasets import load_dataset
# focus only the interesting parquets
data_files = [
r'hanoi_8GB_1Y/node_demand-0-dynamic_output.parquet',
r'hanoi_8GB_1Y/node_demand-1-dynamic_output.parquet',
r'hanoi_8GB_1Y/node_demand-2-dynamic_output.parquet',
]
dataset = load_dataset("rugds/ditec-wdn",data_files=data_files,split='train')
# filter the datasets
historical_dataset = dataset.filter(lambda x: x['scenario_id'] == 0 and 6108 <= x['time_id'] and x['time_id'] < 6131)
calibration_dataset = dataset.filter(lambda x: x['scenario_id'] == 0 and 6132 <= x['time_id'] and x['time_id'] < 7445 and x['time_id'])
# print the values of node 2
print(historical_dataset['2'])
print(calibration_dataset['2'])
B. can you provide a source-faithful bounded export or a documented extraction process with evidence that the protected values at time IDs 7446 and later are not decoded or emitted?
We don't get what you mentioned here. If you cannot find something in the metadata, this means that the dataset generator didn't save it.
1. The authoritative mapping from demand columns 2–32 to Hanoi junction IDs.
The column names should match the node/link names in any network. If you want to check the type of a column name (junction, reservoir, ...), we recommend checking the onamesattribute in the metadata.
2. The exact generator revision and scenario-generation provenance for this published object.
Generator code is on our github. Any configuration, miscellaneous info should be found in the metadata. Using the provided code in the README, you can get them directly from the file containing the "published object".
3. The matching scenario-specific hydraulic input file and its revision or hash.
Unfortunately, INP files were omitted because the dataset already contained most of the relevant information.
4. The release-specific engineering units and meaning of node_demand.
The unit of demand is m^3/s. You can check other units here. In addition, node demand is the "actual" demand that is computed by WNTR/Epanet after the simulation run.
5. The source and output hashes, extraction procedure, and physical storage evidence for any bounded artifact.
Sorry, we don't get what exactly you wants here. Could you please elaborate a bit more?
Regarding the procedure, we assume you are referring to the generation process that is detailed in our paper.
We hope this could help. If you have further question, please let us know.
Cheers,
Huy and Andres