marketdata / README.md
leiyv's picture
Upload README.md with huggingface_hub
28d12aa verified
---
license: mit
task_categories:
- time-series-forecasting
- regression
tags:
- financial
- prediction-market
- polymarket
- time-series
size_categories:
- 10M<n<100M
---
# Polymarket时间序列预测数据集
## 数据集描述
这是一个从Polymarket提取的时间序列预测数据集,用于训练WaveNet模型预测未来市场走势。
### 数据集统计
- **总样本数**: 6,780,504
- **特征维度**: 1,500 (100个时间段 × 15个因子)
- **标签维度**: 10 (预测未来10个时间段)
- **文件大小**: ~12.36 GB
- **格式**: Parquet
### 特征描述
每个样本包含:
- **特征**: 100个历史时间段的15个因子特征 (1,500维)
- **标签**: 未来10个时间段的response_zscore (10维)
- **市场ID**: 标识数据来源的市场
### 15个因子
1. price_std_zscore - 价格标准差
2. price_trend_zscore - 价格趋势
3. price_skew_zscore - 价格偏度
4. price_range_zscore - 价格范围
5. vol_sum_abs_zscore - 成交量总和
6. vol_std_zscore - 成交量标准差
7. vol_flow_zscore - 成交量流
8. vwap_zscore - 成交量加权平均价
9. weighted_volatility_zscore - 加权波动率
10. vol_price_corr_zscore - 量价相关性
11. energy_zscore - 能量
12. flow_ratio_zscore - 流动比率
13. imbalance_intensity_zscore - 不平衡强度
14. vwap_diff_prev_zscore - VWAP差分
15. vwap_slope_ma5_zscore - VWAP斜率
### 数据分割
- **训练集**: 80%
- **验证集**: 10%
- **测试集**: 10%
## 使用方法
### 使用Hugging Face Datasets
```python
from datasets import load_dataset
dataset = load_dataset("chaoleiyv/marketdata", split="train")
```
### 直接使用Parquet文件
```python
import pandas as pd
df = pd.read_parquet("ml_dataset.parquet")
```
## 数据集结构
```
ml_dataset/
├── ml_dataset.parquet # 完整数据集(Parquet格式)
└── README.md # 数据集说明
```
## 引用
如果使用本数据集,请引用:
```bibtex
@dataset{polymarket_ml_dataset,
title={Polymarket时间序列预测数据集},
author={chaoleiyv},
year={2025},
url={https://huggingface.co/datasets/chaoleiyv/marketdata}
}
```
## 许可证
MIT License