Echo0623/PanelTS-ICLR-Review

PanelTS full data release

This release includes all 1,844 CSV files from the public PanelTS source repository at revision bfb751a6735771b5da4e51de2d08fb49a1ad0036, plus 17 canonical selections / 19 Hugging Face configurations. Raw CSV bytes are preserved under PanelTS/; no source data were regenerated. Submission author names are omitted.

Completeness is relative to this pinned upstream snapshot. File counts are not counts of independent scientific datasets. Configurations may overlap (for example, stock groups). coverage.json lists raw files used or unused by the default canonical parsers; all are available in the raw mirror.

Load a selection

from datasets import load_dataset

repo = "Echo0623/PanelTS-ICLR-Review"
data = load_dataset(repo, "covid-20", split="full")
synthetic = load_dataset(repo, "synthetic-y20-x20", split="full", streaming=True)
print(next(iter(synthetic)))

For reproducible results, pass the release commit as revision=.... Large selections support streaming so they need not all fit in memory.

Download the complete original dataset

from huggingface_hub import snapshot_download

snapshot_download("Echo0623/PanelTS-ICLR-Review", repo_type="dataset",
                  allow_patterns=["PanelTS/**", "provenance/**", "coverage.json"],
                  local_dir="panelts-full")

Existing PanelTS API users can read the mirrored source layout directly:

import panelts

ds = panelts.load("stock-healthcare", repo_id="Echo0623/PanelTS-ICLR-Review")
# Or load the downloaded source without network access:
ds = panelts.load("covid-79", data_dir="panelts-full")

Configurations

COVID static attributes are separate covid-20-static and covid-79-static configurations, joined by unit_id. The source provides no separate static-attribute CSV for COVID-320.

Representation and evaluation

Canonical observation tables use time, unit_id, target, and dynamic covariates. Default targets are new COVID cases, Close for market data, and the synthetic target inferred by the SDK. Stock quarterly attributes and exchange-rate macroeconomic attributes use the existing backward alignment. Availability/release lags are not established by alignment; do not assume these attributes were historically available on their observation dates.

full means the complete stored timeline, not a training partition. Configure chronological train/validation/test boundaries and feature availability explicitly. This hosting release does not claim reproduction of paper results. Parser identity, exact source-file hashes, schemas, missing values and round-trip checks are recorded in catalog.json.

The interactive Space remains a small COVID-20 demonstration; the complete dataset is hosted here.

Source attribution and terms

Data are derived from Multiple-Time-Series/PanelTS. The upstream card declares CC BY 4.0, unless otherwise stated for specific source-derived subsets. The unchanged card is retained in provenance/UPSTREAM_README.md. This release preserves that qualification and does not assign new rights over original source data.

Per-selection provenance and responsible-use information are in metadata-review.json and croissant/. Unverified field definitions and original-source details are explicitly identified. Research scores do not establish suitability for medical, public-health or financial decisions.

Integrity and versions

provenance/source-manifest.json records every original CSV's size and SHA-256. SHA256SUMS.json covers every published release file except itself. coverage.json summarizes complete-source coverage. Pin the repository commit for reproducible downloads; historical COVID-20 preview commits remain available.