| --- |
| license: other |
| task_categories: |
| - time-series-forecasting |
| task_ids: |
| - univariate-time-series-forecasting |
| - multivariate-time-series-forecasting |
| annotations_creators: |
| - no-annotation |
| source_datasets: |
| - original |
| tags: |
| - forecasting |
| - benchmark |
| - fev |
| - arxiv:2509.26468 |
| - tsfile |
| - modality:timeseries |
| - timeseries |
| - format:tsfile |
| size_categories: |
| - 10K<n<100K |
| pretty_name: m5 (TsFile format) |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: "**/*.tsfile" |
| modality: |
| - timeseries |
| --- |
| |
| # m5 (TsFile format) |
|
|
| This repository contains time-series forecasting data stored in [Apache TsFile](https://tsfile.apache.org/) format. |
|
|
| ## Summary |
|
|
| - FEV subset: `m5` |
| - Unified source collection: [`autogluon/fev_datasets`](https://huggingface.co/datasets/autogluon/fev_datasets) |
| - Original source: https://www.kaggle.com/competitions/m5-forecasting-accuracy |
| - Paper / citation: [[15]](https://doi.org/10.1016/j.ijforecast.2021.11.013) |
| - Series: 30,490 |
| - Modalities: Time-series |
| - TsFile rows (flattened observations): 503,512,848 |
| - Frequencies: 1D, 1M, 1W |
| - TsFile files: 55 |
| - Time precision: milliseconds (`INT64`). |
|
|
| Licensing and citation requirements follow the original source. This repository does not claim ownership of the original data. |
|
|
| ## Dataset Statistics |
|
|
| | Frequency | Series | Median series length | TsFile rows (observations) | Dynamic columns | Static columns | Data files | |
| |---|---:|---:|---:|---:|---:|---| |
| | 1D | 30,490 | 1,810 | 428,849,460 | 9 | 5 | `1D/1D_1..1D_46.tsfile` (46 shards) | |
| | 1M | 30,490 | 58 | 13,805,685 | 9 | 5 | `1M/1M_1..1M_2.tsfile` (2 shards) | |
| | 1W | 30,490 | 257 | 60,857,703 | 9 | 5 | `1W/1W_1..1W_7.tsfile` (7 shards) | |
|
|
| ## Files |
|
|
| The Hugging Face dataset card YAML points `configs.data_files` to all `*.tsfile` files in this repository. |
|
|
| - `1D/1D_1.tsfile` |
| - `1D/1D_10.tsfile` |
| - `1D/1D_11.tsfile` |
| - `1D/1D_12.tsfile` |
| - `1D/1D_13.tsfile` |
| - `1D/1D_14.tsfile` |
| - `1D/1D_15.tsfile` |
| - `1D/1D_16.tsfile` |
| - `1D/1D_17.tsfile` |
| - `1D/1D_18.tsfile` |
| - `1D/1D_19.tsfile` |
| - `1D/1D_2.tsfile` |
| - `1D/1D_20.tsfile` |
| - `1D/1D_21.tsfile` |
| - `1D/1D_22.tsfile` |
| - `1D/1D_23.tsfile` |
| - `1D/1D_24.tsfile` |
| - `1D/1D_25.tsfile` |
| - `1D/1D_26.tsfile` |
| - `1D/1D_27.tsfile` |
| - ... 35 more `.tsfile` files |
|
|
| ## TsFile Storage Model |
|
|
| - Each original series (`id`) is stored as one TsFile device. |
| - Static covariate columns are stored as TAG columns: `item_id, dept_id, cat_id, store_id, state_id`. |
| - Time-varying targets and dynamic covariates are stored as FIELD measurements. |
| - Source `timestamp` values are mapped to the TsFile `Time` column as millisecond timestamps. |
| - Table name(s): m5_1D, m5_1M, m5_1W. |
| |
| ### Column Schema |
| |
| | Column | Role | TsFile type | |
| |---|---|---| |
| | `Time` | Time column | INT64 | |
| | `id` | TAG (device dimension) | STRING | |
| | `item_id` | TAG (device dimension) | STRING | |
| | `dept_id` | TAG (device dimension) | STRING | |
| | `cat_id` | TAG (device dimension) | STRING | |
| | `store_id` | TAG (device dimension) | STRING | |
| | `state_id` | TAG (device dimension) | STRING | |
| | `target` | FIELD (measurement) | FLOAT | |
| | `snap_CA` | FIELD (measurement) | BOOLEAN | |
| | `snap_TX` | FIELD (measurement) | BOOLEAN | |
| | `snap_WI` | FIELD (measurement) | BOOLEAN | |
| | `sell_price` | FIELD (measurement) | FLOAT | |
| | `event_Cultural` | FIELD (measurement) | BOOLEAN | |
| | `event_National` | FIELD (measurement) | BOOLEAN | |
| | `event_Religious` | FIELD (measurement) | BOOLEAN | |
| | `event_Sporting` | FIELD (measurement) | BOOLEAN | |
|
|
| ## Conversion Notes |
|
|
| - The source FEV format stores each time series as one nested row containing `id`, `timestamp[]`, and target or covariate arrays. |
| - 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. |
| - TAG columns identify the device and static metadata. FIELD columns contain values that change over time. |
| - 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. |
|
|
| ## Reading Example |
|
|
| ```python |
| from tsfile import TsFileReader |
| |
| reader = TsFileReader("1D/1D_1.tsfile") |
| schemas = reader.get_all_table_schemas() |
| # Table name(s): m5_1D, m5_1M, m5_1W |
| ``` |
|
|
