hahuh/PaperBenchPro

---
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。