Dataset Title: Multi-Asset Minute-Level Financial Dataset (May 2021 – April 2022)

Overview:
This dataset contains one-minute frequency financial data collected from 19 assets across five categories:
- Stock Indices and Futures: NASDAQ Composite, S&P 500, DJIA, Nikkei 225, FTSE 100, E-mini Russell 2000, and futures (S&P 500 Futures, DJIA Futures, NASDAQ Futures)
- Commodity Futures: Crude Oil/USD, Silver/USD, Gold/USD
- Cryptocurrencies: BTC/USD, CMC Crypto 200 Index
- Foreign Exchange: GBP/USD, EUR/USD, USD/JPY
- Other Instruments: 10-Year U.S. Treasury Yield, Volatility Index (VIX)

Time Span:
May 16, 2021 – April 20, 2022  
(Only market open days are included)

Data Format:
- Minute-level records
- Each asset has 10-dimensional features per time step:
    [high/end, low/end, open/end, 1 - open/end, 
     m5, m10, m15, m20, m25, m30]
- open and end as the opening and closing prices
- high and low denote the highest and lowest price 
- Moving average (mN) is calculated over N minutes and normalized by the current closing price.


File Format:
All files are in NumPy binary format (.npy) and can be loaded with:
    import numpy as np
    features = np.load('ftse100/features.npy')

File Structure:
- sp500/
    ├── features.npy        # Shape: (3636, 32, 190)
    ├── targets.npy         # Shape: (3636, 24, 2)
    ├── targets_eval.npy    # Shape: (3636, 24, 2)
- ftse100/
    ├── features.npy        # Shape: (4848, 32, 190)
    ├── targets.npy         # Shape: (4848, 24, 2)
    ├── targets_eval.npy    # Shape: (4848, 24, 2)

File Format:
All files are in NumPy binary format (.npy) and can be loaded using:
    import numpy as np
    features = np.load('ftse100/features.npy')

Description:
1. features.npy
   - Shape: (T, N, D)
     - T: Number of time steps (sliding windows) — 3636 for S&P 500, 4848 for FTSE 100
     - N: Input sequence of length = 32
     - D: Feature dimension = 190 (19 financial assets × 10 features each)
2. targets.npy
   - Shape: (T, H, 2) 
   - H: Prediction sequence of length = 24
   - Each label is a binary classification vector:
     - [1, 0] for downward movement
     - [0, 1] for upward movement

3. targets_eval.npy
   - Shape: (T, H, 2) 
   - H: Prediction sequence of length = 24
   - Each entry:
     - [negative_return_abs, 0] if return is negative
     - [0, positive_return] if return is positive
   - Only one of the two elements is non-zero per time step.  
   - Used for calculating return-based performance (e.g., Sharpe ratio).