PXIN/fracture-crypto-l2-enriched
收藏Hugging Face2026-04-29 更新2026-05-03 收录
下载链接:
https://hf-mirror.com/datasets/PXIN/fracture-crypto-l2-enriched
下载链接
链接失效反馈官方服务:
资源简介:
该数据集扩展了Fracture Crypto Dollar Bars,通过从Hyperliquid的原生L2存档中提取的**Level-2订单簿特征**进行增强。每个美元柱都包含了柱关闭时存在的精确微观结构状态,为每个观察提供了流动性、做市商定位和订单流毒性的真实表示。
**源数据(L2):**
- **交易所:** Hyperliquid(原生L1永续DEX)
- **存档:** `s3://hyperliquid-archive/market_data/`
- **交易对:** BTC, ETH, SOL, BNB, HYPE, TRX
- **日期范围:** 2025年9月 — 2026年3月(7个月,覆盖约92%)
- **原始模式:** 20级订单簿快照,每级包含`px`, `sz`, `n`
**源数据(交易):**
- **交易所:** Binance Futures(USD-M永续合约)
- **交易对:** BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, HYPEUSDT, TRXUSDT
- **日期范围:** 2025年9月 — 2026年3月
该数据集结合了美元柱(基于活动采样,标准化波动率制度)和L2快照(基于状态的特征,捕捉瞬时流动性环境),生成的特征向量中每个观察都携带了**宏观活动上下文**(多少交易量推动了价格)和**微观流动性上下文**(发生时订单簿的状态)。
This dataset extends the [Fracture Crypto Dollar Bars](https://huggingface.co/datasets/PXIN/fracture-crypto-dollar-bars) with **Level-2 order book features** extracted from Hyperliquids native L2 archive. Every dollar bar is enriched with the exact microstructure state that existed at the moment the bar closed — providing a ground-truth representation of liquidity, market maker positioning, and order flow toxicity for each observation.
**Source Data (L2):**
- **Exchange:** Hyperliquid (native L1 perp DEX)
- **Archive:** `s3://hyperliquid-archive/market_data/`
- **Pairs:** BTC, ETH, SOL, BNB, HYPE, TRX
- **Date Range:** September 2025 — March 2026 (7 months, ~92% coverage)
- **Raw Schema:** 20-level order book snapshots with `px`, `sz`, `n` per level
**Source Data (Trades):**
- **Exchange:** Binance Futures (USD-M Perpetuals)
- **Pairs:** BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, HYPEUSDT, TRXUSDT
- **Date Range:** September 2025 — March 2026
The dataset combines two orthogonal but complementary sampling frameworks:
- **Dollar bars** (López de Prado, 2018) — activity-based sampling that normalizes for volatility regime
- **L2 snapshots** — state-based features that capture the instantaneous liquidity environment
The result is a feature vector where each observation carries both the **macro activity context** (how much volume moved the price) and the **micro liquidity context** (what the book looked like when it happened).
提供机构:
PXIN



