crypto_model

Crypto Price Predictor V8

A high-performance LSTM-based cryptocurrency price prediction model with bias correction.

Model Details

Architecture

• Type: Bidirectional LSTM
• Layers: 2 stacked LSTM layers
• Hidden Size: 64 (auto-detected per model)
• Input Features: 44 technical indicators
• Output: Next hour price prediction

Supported Cryptocurrencies

BTC, ETH, SOL, BNB, XRP, ADA, DOT, LINK, MATIC, AVAX, FTM, NEAR, ATOM, ARB, OP, LTC, DOGE, UNI, SHIB, PEPE

Performance

• Average MAPE: < 0.05%
• Average MAE: < 50 USD (varies by price)
• Direction Accuracy: ~65-75%

Usage

Quick Start

from bot_predictor import BotPredictor

# Initialize
bot = BotPredictor()

# Get prediction
prediction = bot.predict('BTC')
print(f"Next Hour Price: ${prediction['corrected_price']:.2f}")
print(f"Direction: {prediction['direction']}")
print(f"Confidence: {prediction['confidence']*100:.1f}%")

Installation

pip install torch torchvision torchaudio
pip install scikit-learn pandas numpy ccxt

Models

All models are in PyTorch format (.pth files). Download all models to use the full suite.

Bias Correction

Each model includes an automatic bias correction value to account for training/test distribution differences.

File: bias_corrections_v8.json

Technical Indicators (44 total)

• RSI (14, 21)
• MACD + Signal + Histogram
• Bollinger Bands (20, 2)
• ATR (14)
• CCI (20)
• Momentum (10)
• SMA (5, 10, 20, 50)
• EMA (12, 26)
• Volume Ratio
• OHLC-derived features

Training Details

• Data: 1000 hourly candles per symbol
• Train/Val/Test Split: 80/10/10
• Optimizer: Adam (LR=0.005)
• Loss: MSE
• Batch Size: 64
• Epochs: 150 (with early stopping)
• Dropout: 0.3

Requirements

torch>=2.0.0
torchvision
pandas>=1.5.0
numpy>=1.23.0
scikit-learn>=1.2.0
ccxt>=2.0.0
huggingface_hub>=0.16.0
python-dotenv>=1.0.0

License

MIT License

Citation

@software{crypto_predictor_v8,
  title={Crypto Price Predictor V8},
  author={Your Name},
  year={2025},
  url={https://huggingface.co/caizongxun/crypto-price-predictor-v8}
}

Disclaimer

These models are for educational and research purposes only. Do not use for actual trading without thorough validation.

Support

For issues and questions, please refer to the GitHub repository.