THULab/ercot
ercot (TsFile format) This repository contains time-series forecasting data stored in Apache TsFile format. Summary FEV subset: ercot Unified source collection: autogluon/fev_datasets Original source: https://github.com/ourownstory/neuralprophet-data/tree/main/datasets_raw/energy Series: 8 Modalities: Time-series TsFile rows (flattened observations): 1,299,648 Frequencies: 1D, 1H, 1M, 1W TsFile files: 5 Time precision: milliseconds (INT64). Licensing and… See the full description on the dataset page: https://huggingface.co/datasets/THULab/ercot.
064
1---2license: other3task_categories:4- time-series-forecasting5task_ids:6- univariate-time-series-forecasting7- multivariate-time-series-forecasting8annotations_creators:9- no-annotation10source_datasets:11- original12tags:13- forecasting14- benchmark15- fev16- arxiv:2509.2646817- tsfile18- modality:timeseries19- timeseries20- format:tsfile21size_categories:22- n<1K23pretty_name: ercot (TsFile format)24configs:25- config_name: default26 data_files:27 - split: train28 path: "**/*.tsfile"29modality:30- timeseries31---32 33# ercot (TsFile format)34 35This repository contains time-series forecasting data stored in [Apache TsFile](https://tsfile.apache.org/) format.36 37## Summary38 39- FEV subset: `ercot`40- Unified source collection: [`autogluon/fev_datasets`](https://huggingface.co/datasets/autogluon/fev_datasets)41- Original source: https://github.com/ourownstory/neuralprophet-data/tree/main/datasets_raw/energy42- Series: 843- Modalities: Time-series44- TsFile rows (flattened observations): 1,299,64845- Frequencies: 1D, 1H, 1M, 1W46- TsFile files: 547- Time precision: milliseconds (`INT64`).48 49Licensing and citation requirements follow the original source. This repository does not claim ownership of the original data.50 51## Dataset Statistics52 53| Frequency | Series | Median series length | TsFile rows (observations) | Dynamic columns | Static columns | Data files |54|---|---:|---:|---:|---:|---:|---|55| 1D | 8 | 6,452 | 51,616 | 1 | 0 | `1D/1D.tsfile` |56| 1H | 8 | 154,872 | 1,238,976 | 1 | 0 | `1H/1H_1..1H_2.tsfile` (2 shards) |57| 1M | 8 | 211 | 1,688 | 1 | 0 | `1M/1M.tsfile` |58| 1W | 8 | 921 | 7,368 | 1 | 0 | `1W/1W.tsfile` |59 60## Files61 62The Hugging Face dataset card YAML points `configs.data_files` to all `*.tsfile` files in this repository.63 64- `1D/1D.tsfile`65- `1H/1H_1.tsfile`66- `1H/1H_2.tsfile`67- `1M/1M.tsfile`68- `1W/1W.tsfile`69 70## TsFile Storage Model71 72- Each original series (`id`) is stored as one TsFile device.73- Time-varying targets and dynamic covariates are stored as FIELD measurements.74- Source `timestamp` values are mapped to the TsFile `Time` column as millisecond timestamps.75- Table name(s): ercot_1D, ercot_1H, ercot_1M, ercot_1W.76 77### Column Schema78 79| Column | Role | TsFile type |80|---|---|---|81| `Time` | Time column | INT64 |82| `id` | TAG (device dimension) | STRING |83| `target` | FIELD (measurement) | FLOAT |84 85## Conversion Notes86 87- The source FEV format stores each time series as one nested row containing `id`, `timestamp[]`, and target or covariate arrays.88- The TsFile conversion flattens those nested arrays into long rows. Therefore, the `TsFile rows` values above correspond to the number of timestamped observations after flattening.89- TAG columns identify the device and static metadata. FIELD columns contain values that change over time.90- Large logical tables may be split into multiple `.tsfile` shards such as `<name>_1.tsfile`, `<name>_2.tsfile`, and so on. Shards listed for the same frequency belong to the same logical table.91 92## Reading Example93 94```python95from tsfile import TsFileReader96 97reader = TsFileReader("1D/1D.tsfile")98schemas = reader.get_all_table_schemas()99# Table name(s): ercot_1D, ercot_1H, ercot_1M, ercot_1W100```101