Datasets:
Download README.md from THULab/ts-arena-weather: direct link, hf CLI and curl.
- Browser
- Download file 5.23 kB
-
https://huggingface.co/datasets/THULab/ts-arena-weather/resolve/main/README.md
- Command line
-
hf download hf://datasets/THULab/ts-arena-weather/README.md
-
curl -L -o README.md https://huggingface.co/datasets/THULab/ts-arena-weather/resolve/main/README.md
license: mit
task_categories:
- time-series-forecasting
tags:
- weather
- autoformer
- ts-arena
- forecasting
- long-horizon
- tsfile
- format:tsfile
- timeseries
pretty_name: Weather (Max Planck, Autoformer) — TsFile format
size_categories:
- 10K<n<100K
Weather (Max Planck Institute, Autoformer) — TsFile format
This repository is a conversion to Apache TsFile format of the Weather time-series dataset, originating from the TS Arena benchmark suite.
- Original dataset (Hugging Face): ts-arena/weather
- Original data source: Autoformer Dataset Collection (Max Planck Institute for Biogeochemistry weather station, 2020)
- License: MIT
Dataset Description
Multivariate weather time series from the Max Planck weather station in 2020, sampled every 10 minutes, with 52,696 time points and 21 weather features — a common benchmark for long-sequence time-series forecasting (LSTF). The dataset is physically split into train / validation / test (70% / 10% / 20%).
| File | split | rows |
|---|---|---|
data/train.tsfile |
train | 36,887 |
data/val.tsfile |
validation | 5,269 |
data/test.tsfile |
test | 10,540 |
⚠️ Data is standardized
The feature values in the original dataset are already standardized (zero mean / unit variance, scaler fitted on train only). This repository keeps the standardized values verbatim and does not de-standardize. To recover the original physical quantities, use scaler_params.json at the repo root (containing per-column mean / std / feature_names):
original = standardized * std + mean
Column meanings
| Column | Meaning | Type |
|---|---|---|
Time |
time index (INT64, from source timestamp_idx, 0/1/2…; the source has no real timestamps) |
time column |
p_mbar |
pressure p (mbar) | FLOAT |
T_degC |
air temperature T (°C) | FLOAT |
Tpot_K |
potential temperature Tpot (K) | FLOAT |
Tdew_degC |
dew point temperature Tdew (°C) | FLOAT |
rh_pct |
relative humidity rh (%) | FLOAT |
VPmax_mbar |
saturation vapor pressure VPmax (mbar) | FLOAT |
VPact_mbar |
actual vapor pressure VPact (mbar) | FLOAT |
VPdef_mbar |
vapor pressure deficit VPdef (mbar) | FLOAT |
sh_g_kg |
specific humidity sh (g/kg) | FLOAT |
H2OC_mmol_mol |
water-vapor concentration H2OC (mmol/mol) | FLOAT |
rho_g_m3 |
air density rho (g/m³) | FLOAT |
wv_m_s |
wind speed wv (m/s) | FLOAT |
max_wv_m_s |
max. wind speed wv (m/s) | FLOAT |
wd_deg |
wind direction wd (deg) | FLOAT |
rain_mm |
rainfall rain (mm) | FLOAT |
raining_s |
rainfall duration raining (s) | FLOAT |
SWDR_W_m2 |
shortwave downward radiation SWDR (W/m²) | FLOAT |
PAR_umol_m2_s |
photosynthetically active radiation PAR (µmol/m²/s) | FLOAT |
max_PAR_umol_m2_s |
max. PAR (µmol/m²/s) | FLOAT |
Tlog_degC |
recorded temperature Tlog (°C) | FLOAT |
OT |
forecast target (Output Target) | FLOAT |
Repository structure
ts-arena-weather/
├── README.md # this file
├── scaler_params.json # standardization-reversal params (mean / std / feature_names)
└── data/
├── train.tsfile
├── val.tsfile
└── test.tsfile
Conversion Notes
- Each split is converted to its own
.tsfile, not merged (preserving the original train/val/test split). - Time column: from the source
timestamp_idx(int64, 0/1/2…) as INT64. The source data has no real timestamps. - Standardized values kept verbatim: no de-standardization; reversal params are in
scaler_params.json. - Field columns: all 21 weather features are stored as single-precision FLOAT, with column names cleaned into TsFile-safe identifiers (units kept in the names). The
²/µin the original column names were corrupted into U+FFFD and are restored by physical meaning toW_m2/umol_m2_s. - No data columns dropped: all 22 columns are kept (21 features +
timestamp_idx→Time). - No TAG columns: each split is an independent time series.
Reading example
from tsfile import TsFileReader
reader = TsFileReader("data/train.tsfile")
schemas = reader.get_all_table_schemas()
tname = next(iter(schemas))
field_cols = [c.get_column_name() for c in schemas[tname].get_columns()]
with reader.query_table(tname, field_cols, batch_size=65536) as rs:
while (batch := rs.read_arrow_batch()) is not None:
df = batch.to_pandas()
print(df.head())
break
Source and citation
Data from the Max Planck Institute for Biogeochemistry weather station, curated by the Autoformer dataset collection and published (standardized) by TS Arena.
@inproceedings{wu2021autoformer,
title={Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting},
author={Wu, Haixu and Xu, Jiehui and Wang, Jianmin and Long, Mingsheng},
booktitle={Advances in Neural Information Processing Systems},
year={2021}
}