AcyLa/crypto-predictive-model

🏦 Crypto Predictive Model — ML Ensemble for Crypto Price Prediction

A production-grade machine learning pipeline for cryptocurrency price prediction using 15 years of Yahoo Finance historical data, engineered for integration with high-frequency trading systems.

🚀 Quick Start

pip install yfinance pandas numpy scikit-learn lightgbm ta torch pyarrow joblib
git clone https://huggingface.co/AcyLa/crypto-predictive-model
cd crypto-predictive-model
python train.py --ticker BTC-USD
python predict.py --ticker BTC-USD

🏗 Architecture

Data (yfinance) → Features (207 indicators) → Labels (multi-horizon) → Ensemble (LightGBM+LSTM) → Prediction API
                                                          ↑
                                          Walk-Forward Backtester (purged CV)

📊 Components

Component	Description
Data	Downloads up to 15 years of OHLCV from Yahoo Finance (BTC since 2014)
Features	207 causal technical indicators: trend, momentum, volatility, volume, price structure, regime, calendar
Labels	Multi-horizon (1d, 3d, 7d, 14d, 30d) regression + classification + triple-barrier
Models	Ensemble of LightGBM (gradient boosting) + LSTM with attention pooling
Backtest	Walk-forward cross-validation with purging — no look-ahead bias
Prediction	Real-time multi-horizon signals with consensus voting

🔧 HFT Integration

from predict import CryptoPredictor

predictor = CryptoPredictor(ticker="BTC-USD")
result = predictor.get_multi_horizon(use_lstm=False)  # sub-ms for HFT
print(f"Consensus: {result['consensus']} ({result['consensus_confidence']:.0%})")

See hft_integration.py for a complete trading loop with stop-loss, take-profit, and position management.

Mode	Latency	Use Case
LightGBM only	< 1ms	HFT, tick-level
LightGBM + LSTM (CPU)	~6-50ms	Swing trading

📁 Files

├── config.py          # All hyperparameters
├── data_loader.py     # Yahoo Finance download
├── features.py        # 207 technical indicators
├── labels.py          # Multi-horizon labels + triple-barrier
├── models.py          # LightGBM + LSTM ensemble
├── backtest.py        # Walk-forward backtesting engine
├── train.py           # Main training pipeline
├── predict.py         # Live prediction API + CLI
├── hft_integration.py # HFT trading loop example
└── models/BTC-USD/    # Trained models

⚠️ Disclaimer

Research tool, not financial advice. Always backtest with transaction costs and never risk more than you can afford to lose.

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "AcyLa/crypto-predictive-model"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

For non-causal architectures, replace AutoModelForCausalLM with the appropriate AutoModel class.