danguo-research's picture
Upload feature version 3.1.0
79d5308 verified
|
Raw History Blame Contribute Delete
7.54 kB
metadata
pretty_name: QuantFeatures Hyperliquid Live 10-Minute Features (Sample)
license: other
license_name: quantfeatures-sample-evaluation
license_link: LICENSE.md
language:
  - en
tags:
  - finance
  - crypto
  - cryptocurrency
  - market-microstructure
  - time-series
  - features
  - hyperliquid
  - real-time
task_categories:
  - time-series-forecasting
size_categories:
  - 10K<n<100K
configs:
  - config_name: TradeFlow
    default: true
    data_files:
      - split: perp
        path: data/10min/TradeFlow/perp.parquet
  - config_name: TradeTick
    data_files:
      - split: perp
        path: data/10min/TradeTick/perp.parquet
  - config_name: TradeCovar
    data_files:
      - split: perp
        path: data/10min/TradeCovar/perp.parquet
  - config_name: TradeCovarPlus
    data_files:
      - split: perp
        path: data/10min/TradeCovarPlus/perp.parquet
  - config_name: TradeStats
    data_files:
      - split: perp
        path: data/10min/TradeStats/perp.parquet
  - config_name: TradeGroup
    data_files:
      - split: perp
        path: data/10min/TradeGroup/perp.parquet
  - config_name: TradeGroupPlus
    data_files:
      - split: perp
        path: data/10min/TradeGroupPlus/perp.parquet
  - config_name: TradeLarge
    data_files:
      - split: perp
        path: data/10min/TradeLarge/perp.parquet
  - config_name: TradeGradient
    data_files:
      - split: perp
        path: data/10min/TradeGradient/perp.parquet
  - config_name: TradeImpact
    data_files:
      - split: perp
        path: data/10min/TradeImpact/perp.parquet
  - config_name: QuoteBasic
    data_files:
      - split: perp
        path: data/10min/QuoteBasic/perp.parquet
  - config_name: FundingRate
    data_files:
      - split: perp
        path: data/10min/FundingRate/perp.parquet
  - config_name: OpenIntBasic
    data_files:
      - split: perp
        path: data/10min/OpenIntBasic/perp.parquet
  - config_name: OpenIntPlus
    data_files:
      - split: perp
        path: data/10min/OpenIntPlus/perp.parquet
  - config_name: OpenIntGradient
    data_files:
      - split: perp
        path: data/10min/OpenIntGradient/perp.parquet
extra_gated_heading: Request access to the QuantFeatures sample
extra_gated_prompt: >-
  This sample is provided for evaluating QuantFeatures data products under the
  QuantFeatures Sample Evaluation License (see LICENSE.md). Redistribution,
  resale, and production or live-trading use are not permitted without a
  separate agreement.
extra_gated_fields:
  Full name: text
  Company or institution: text
  Work email: text
  Intended use:
    type: select
    options:
      - Evaluating data for research
      - Evaluating data for a commercial license
      - Academic research
      - label: Other
        value: other
  I agree to the QuantFeatures Sample Evaluation License and will not redistribute the data: checkbox
extra_gated_button_content: Agree and request access

QuantFeatures Hyperliquid Live 10-Minute Features (Sample)

10-minute features for Hyperliquid perpetuals, computed as each interval ends: trade activity and flow, trade distributions, top of book, funding rate and open interest. Each update replaces the files with the last 7 UTC days, today included. For every day of 2026, hourly and 10-minute, see QuantFeatures/hyperliquid-features.

Attribute Value
Instruments BCH, BNB, BTC, ETH, HYPE, XRP (quoted in USDC; the quote column reads usdt)
Interval 10 minutes
Window The last 7 UTC days, now 2026-11-05 to 2026-11-11 11:30 UTC (last interval start)
Update Every 10 minutes, a few minutes after each interval ends (Kaggle takes longer to process each version)

Feature version 3.1.0; 64,780 rows across 15 files.

Source

Hyperliquid's public websocket trade, best bid/offer and asset context (funding, open interest, mark price) streams, recorded by QuantFeatures with exchange timestamps. No third-party data vendor is involved. Intervals are cut by exchange time. Each interval is computed from the state saved at the end of the previous one.

The same intervals are computed again from the full day's recording the next day (the 10-minute configs of QuantFeatures/hyperliquid-features), and the two are compared every day. Trade modules match exactly; quote, funding and open-interest modules differ slightly in a few intervals a day. The next-day files are the reference.

Load

import pandas as pd
perp = pd.read_parquet("hf://datasets/QuantFeatures/hyperliquid-live-features/data/10min/TradeFlow/perp.parquet")

Modules (configs)

Config Family Input Columns Rows
TradeFlow Trade activity Exchange trades 45 4,504
TradeTick Extended trade activity Exchange trades 33 4,504
TradeCovar Trade relationships Exchange trades 30 4,504
TradeCovarPlus Extended trade relationships Exchange trades 54 4,504
TradeStats Trade distribution Exchange trades 24 4,504
TradeGroup Trade grouping Exchange trades 123 4,504
TradeGroupPlus Grouped trade distribution Exchange trades 124 4,504
TradeLarge Trade tails Exchange trades 123 4,504
TradeGradient Trade dynamics Exchange trades 42 4,504
TradeImpact Trade dynamics Exchange trades 42 4,504
QuoteBasic Top of book Best-bid/offer quotes 71 3,948
FundingRate Funding Derivative ticker feed 33 3,948
OpenIntBasic Open interest Derivative ticker feed 33 3,948
OpenIntPlus Extended open interest Derivative ticker feed 45 3,948
OpenIntGradient Open-interest dynamics Derivative ticker feed 39 3,948

Full field definitions are in the Feature Catalog.

Record semantics

  • Each row is one instrument in one half-open UTC interval [start_time_utc, end_time_utc).
  • Rows are sorted by start_time_utc, exch, then base.
  • An interval's information is complete only at end_time_utc. Use end_time_utc as the availability time.
  • NaN means unavailable or not validly computable, never zero. Integer counts are null only in masked intervals.
  • Prices are in USDC, volumes in the base asset, and _bps fields in basis points.

manifest.json lists row counts, the window and a SHA-256 checksum for every file.

Masked intervals

When our recording is interrupted for 60 seconds or more, every interval that overlaps the interruption has all feature values NaN (counts null); the rows are kept. An interruption is known only after the recorder reconnects, so an interval can be masked in a later update. In the current window:

  • 2026-11-05, intervals starting 00:50, 01:00, 01:10, 01:20, 01:30 UTC: recording interrupted 00:57:40 to 01:32:27 UTC
  • 2026-11-06, intervals starting 00:50 UTC: recording interrupted 00:50:54 to 00:54:42 UTC
  • 2026-11-07, intervals starting 08:10, 08:20 UTC: recording interrupted 08:18:12 to 08:20:17 UTC

License

QuantFeatures Sample Evaluation License (LICENSE.md): you may use this sample internally to sample and review QuantFeatures data products. Redistribution, resale, production use, reconstructing underlying market data, and reverse-engineering the feature methods are not permitted without a separate agreement. No raw exchange or vendor data is included. Not investment advice.

Contact

More instruments, other intervals, or delivery to your own storage: QuantFeatures@gmail.com