Rewrite dataset card in English; add tsfile tag

README.md

@@ -1,46 +1,51 @@
-# fastest (TsFile format)
-
-Converted from the HuggingFace dataset [`FaalSa/fastest`](https://huggingface.co/datasets/FaalSa/fastest) into Apache TsFile format.
-
-## Layout
+---
+language:
+- en
+license: apache-2.0
+task_categories:
+- time-series-forecasting
+tags:
+- tsfile
+- time-series
+- forecasting
+pretty_name: fastest
+size_categories:
+- 10K<n<100K
+---
+
+# fastest (TsFile)
+
+Apache TsFile version of
+[`FaalSa/fastest`](https://huggingface.co/datasets/FaalSa/fastest).
+
+## Overview
+
+A univariate time-series forecasting dataset of **25 series**, provided in the standard
+GluonTS-style train / validation / test layout. The validation and test splits extend the
+train series by one and two prediction horizons respectively (validation = train + 120
+points, test = train + 240 points).
 
 ```
 train/      faalsa_fastest_train.tsfile        25 series × 4320 points = 108,000 rows
 validation/ faalsa_fastest_validation.tsfile   25 series × 4440 points = 111,000 rows
-test/       faalsa_fastest_test.tsfile          25 series × 4560 points = 114,000 rows
+test/       faalsa_fastest_test.tsfile         25 series × 4560 points = 114,000 rows
 ```
 
-The original train/validation/test split is preserved — each split is a separate `.tsfile` (not merged).
-validation = train + 120 points, test = train + 240 points (one and two prediction horizons).
-
-## Schema
-
-Each `.tsfile` contains one table with:
+The original train / validation / test split is preserved — each split is a separate TsFile
+(not merged).
 
-| column            | role        | type   | notes                                            |
-|-------------------|-------------|--------|--------------------------------------------------|
-| `item_id`         | tag         | STRING | series id (one device per series)                |
-| `feat_cat`        | tag         | STRING | from `feat_static_cat[0]`, e.g. `cat_0`, `cat_1` |
-| `Time`            | time column | INT64  | synthesized hourly timestamps                    |
-| `target`          | field       | FLOAT  | the series values                                |
+## Schema (TsFile structure)
 
-25 devices per file (one per `item_id`), each carrying the full series for that item.
+- **Time** (INT64, milliseconds) — built from the per-series `start` timestamp and position.
+- **item_id** (TAG) — the series identity (25 devices).
+- **target** (FIELD, FLOAT) — the observed value.
 
-## Conversion notes
+## Usage
 
-- Each source row is one full time-series; the array column `target` was flattened so that
-  every point becomes one row, with `Time` synthesized from `start` (2021-01-01 00:00:00).
-- **Sampling frequency**: the source dataset does **not** declare a frequency. It was inferred
-  as **hourly** from row-count evidence (train = 4320 points = exactly 180 days; the prediction
-  horizon between splits is 120 points). If the true frequency differs, only the `Time` axis
-  spacing changes — the values and tags are unaffected.
-- `feat_static_cat` (a single-element array per series) was scalarized into the string tag
-  `feat_cat` (e.g. `[0] -> cat_0`).
-- **No columns were dropped**: all 4 source columns are preserved
-  (`start` → `Time`, `target` → field, `item_id` + `feat_static_cat` → tags).
+Read the `.tsfile` files with the Apache TsFile Java or Python SDK.
 
-## Source
+## Source & license
 
-- Original dataset: [FaalSa/fastest](https://huggingface.co/datasets/FaalSa/fastest)
+- Original dataset: https://huggingface.co/datasets/FaalSa/fastest
 - Author: FaalSa
-- Task: time-series forecasting
+- License: apache-2.0