blobxiaoyao/MagSeven-ML-Ready-High-Frequency-Anomaly-Dataset-January-2026-sample

MagSeven High-Frequency Anomaly Dataset (Sample)

This repository provides a public sample dataset extracted from a larger, production-ready financial time series dataset.

The goal of this sample is to demonstrate:

• Data structure
• Feature engineering style
• Cleanliness and usability for modeling

---

📦 Dataset Overview

This dataset contains normalized and relative time-series features derived from raw OHLCV market data.

Instead of absolute prices, the data focuses on relative movements anchored to an open price, making it more suitable for:

• Machine learning models
• Cross-asset generalization
• Regime-agnostic pattern discovery

---

📊 Event-Level Data Schema

Each JSON object corresponds to a single detected anomaly event and follows the schema below:

{
  "ticker": "AAPL",
  "timestamp_utc": "2026-01-06 14:30:00+00:00",
  "metrics": {
    "market_phase": "OPEN_AUCTION",
    "event_type": "MARKET_LIQUIDITY",
    "anomaly_strength": "Level_2",
    "volume_z_score": 6.32,
    "volatility_z_score": 2.15,
    "price_change_from_window_open": 0.003,
    "outcomes": {
      "future_return_30m": 0.0045,
      "max_upside_30m": 0.008,
      "max_drawdown_30m": -0.001,
      "label": "BULLISH"
    }
  },
  "data_1m": [ ... ],
  "data_5m": [ ... ]
}

🧱 Field Overview

Top-Level Fields

• ticker: Stock symbol associated with the anomaly event.
• timestamp_utc: UTC timestamp corresponding to the anchor (trigger) bar.
• metrics: Event-level summary statistics, detection logic outputs, and forward-looking outcome labels.
• data_1m: High-resolution 1-minute microstructure window centered around the anomaly event.
• data_5m: Lower-frequency 5-minute context window capturing short-term trend behavior.

🛠 Usage Notes

• All price-related fields are normalized to protect data source agreements and emphasize relative price action dynamics.
• Absolute price levels are intentionally excluded.
• The dataset is optimized for direct ingestion into ML pipelines without additional preprocessing.

🧪 Example: Loading the Dataset in Python

import pandas as pd

# Load anomaly data for a single ticker
df = pd.read_json("AAPL_anomaly_package.jsonl", lines=True)

# Inspect event-level metrics
print(df.iloc[0]["metrics"])

# Convert the 1-minute window into a DataFrame
event_1m = pd.DataFrame(df.iloc[0]["data_1m"])
print(event_1m.head())

📚 Feature Dictionary (data_1m & data_5m)

Both data_1m (microstructure) and data_5m (trend context) arrays share the same feature definitions.

Feature	Description
Datetime	Timestamp of the bar (UTC).
is_anchor	True if this bar corresponds to the triggered anomaly event; False for surrounding context bars.
Rel_Open / Rel_High / Rel_Low / Rel_Close	Price relative to the open price of the first bar in the window. Rel_Price = (Price - Window_Start_Open) / Window_Start_Open
Price_Velocity	Percentage change in Close price from the previous bar.
Price_Acceleration	Change in Price Velocity (second derivative of price).
Log_Return	Logarithmic return of the Close price.
Volume_Ratio	Current volume divided by the trailing moving average volume. Measures relative volume intensity.
Money_Flow_Ratio	Current money flow (Close × Volume) divided by its trailing moving average.
Volume_Momentum	First derivative of volume (change in volume from the previous bar).
Volatility_Z_Score	Standardized score of the High–Low range relative to recent history.
Vol_Z_Change	Change in Volatility Z-Score from the previous bar.
Bar_Intensity	(Close - Open) / (High - Low). Measures directional conviction (range: -1.0 to 1.0).
Distance_to_MA20	Percentage distance of the Close price from the 20-period moving average.

---

🧠 Intended Use Cases

This dataset format is suitable for:

• Time-series forecasting
• Anomaly detection
• Pattern discovery
• Quantitative research
• Feature engineering benchmarks
• ML / DL pipelines (LSTM, Transformer, etc.)

---

🧪 About This Sample

This Hugging Face repository contains only a small subset of the full dataset.

It is intended for:

• Inspection
• Experimentation
• Pipeline testing

The full dataset includes:

• Larger time coverage
• Multiple instruments
• Extended metadata
• Ready-to-train splits

---

🔗 Full Dataset (Gumroad)

The complete dataset is available for purchase on Gumroad:

👉 Get the full dataset on Gumroad

---

📄 License

This sample dataset is released under the CC-BY-4.0 License.

You are free to use it for research and experimentation, with attribution.

---

📬 Contact

If you have questions, feedback, or custom data requests, feel free to reach out via Gumroad.

---

📬 Contact & Support

If you have any questions about this dataset, licensing, or access to the full version, feel free to reach out:

📧 Email: quantalpha.global@gmail.com

Please note that this email is intended for dataset-related inquiries only.
We aim to respond within 1–2 business days.