Celsia/HPEC2026

HPEC 2026 — HS6 US Trade Demand-Forecasting Lattice

Dataset for the IEEE HPEC 2026 paper "Are Multidimensional Models Worth it in Demand Forecasting?" (Paper ID 372). Built from the U.S. Census Bureau's bulk port-level HS6 merchandise trade files (public domain), it provides a dense commodity × state × flow lattice for benchmarking flat vs. explicitly multidimensional sequence models (RNNs, Transformers, S4/S4ND, Mamba-2/3, CaFA-style factorized attention).

Code (models, training sweeps, lattice builder): https://github.com/triadastra/Multidimensional-Long-Series-Forecasting

Contents

raw/                          384 Census bulk ZIPs — port-level HS6 exports + imports, 2010–2025
raw/reference/                Census codebooks and fixed-width record layouts
processed/census_lattice_9ch.npz    main benchmark lattice (9 channels, ~219 MB)
processed/census_lattice_9ch.json   sidecar metadata (REQUIRED — see below)
processed/census_lattice.npz        earlier 2-channel build (~64 MB)
processed/census_lattice.json       its sidecar

Processed lattice (census_lattice_9ch.npz)

• Grid: 1,263 HS6 commodities × 14 port-states × 2 flows (export/import), 192 monthly steps (2010–2025); cells kept at ≥80% density → 28,292 active series.
• Channels (9): aggregate + air/containerized/bulk/land value, and aggregate + air/containerized/bulk weight; all log1p then per-series MinMax fit on the training split only.
• Sidecar JSON is load-bearing: the .npz carries no metadata. Target channels, axis labels, and split boundaries live in the matching .json — always load it alongside (target_channels = [0, 5] for the 9-channel build; a script reading the npz alone will silently pick wrong defaults).
• Built by scripts/build_census_lattice.py in the GitHub repo (HS4 dev variant via --level 4).

Raw files

230-character fixed-width records: HS6 (Schedule B) + country (Schedule C) + district/port (Schedule D → state) + YYYYMM + value/weight fields. Schedule B / HS codes were re-versioned in 2012, 2017, and 2022 — match each year to its vintage code list (layouts and vintage code tables in raw/reference/).

License & provenance

Source data is produced by the U.S. Census Bureau and is in the public domain (17 U.S.C. § 105). This repackaging adds no restrictions.

Citation

@inproceedings{fan2026multidim,
  title     = {Are Multidimensional Models Worth it in Demand Forecasting?},
  author    = {Fan, Cheng-Jui and Aristov, Nikolay and Dugundji, Elenna},
  booktitle = {IEEE High Performance Extreme Computing Conference (HPEC)},
  year      = {2026}
}