README.md

---
license: mit
task_categories:
  - time-series-forecasting
  - time-series-classification
tags:
  - time-series
  - benchmark
  - forecasting
  - anomaly-detection
size_categories:
  - 1M<n<10M
---

# PaperBenchPro Time Series Datasets

时序分析基准数据集合集，按任务类型组织。

## 📂 目录结构

```
dataset_by_task/
├── anomaly_detection/      # 异常检测
│   ├── MSL/
│   ├── PSM/
│   ├── SMAP/
│   ├── SMD/
│   └── SWaT/
├── classification/         # 时序分类
│   ├── EthanolConcentration/
│   ├── FaceDetection/
│   ├── Handwriting/
│   ├── Heartbeat/
│   ├── JapaneseVowels/
│   ├── PEMS-SF/
│   ├── SelfRegulationSCP1/
│   ├── SelfRegulationSCP2/
│   ├── SpokenArabicDigits/
│   └── UWaveGestureLibrary/
├── imputation/             # 数据填补
│   ├── electricity/
│   ├── ETT-small/
│   └── weather/
├── long_term_forecast/     # 长期预测
│   ├── electricity/
│   ├── ETT-small/
│   ├── exchange_rate/
│   ├── illness/
│   ├── traffic/
│   └── weather/
└── short_term_forecast/    # 短期预测
    └── m4/
```

## 📊 任务说明

| 任务类型 | 描述 | 数据集数量 |
|---------|------|-----------|
| anomaly_detection | 时序异常检测 | 5 |
| classification | 时序分类 | 10 |
| imputation | 缺失值填补 | 6 |
| long_term_forecast | 长期预测 (96/192/336/720步) | 9 |
| short_term_forecast | 短期预测 (M4竞赛) | 1 |

## 🚀 使用方式

```python
from huggingface_hub import snapshot_download

# 下载全部数据
snapshot_download(repo_id="hahuh/PaperBenchPro", repo_type="dataset")

# 只下载特定任务
snapshot_download(
    repo_id="hahuh/PaperBenchPro",
    repo_type="dataset",
    allow_patterns="anomaly_detection/*"
)
```

## 📝 引用

如果使用本数据集，请引用相关论文。