matthewchung74/stock-algo-nvda-model

Exported Trading Models

This package contains trained XGBoost models for stock price prediction, hosted on Hugging Face.

Models Included

1 trained models:

• NVDA: 5-minute bar prediction model

Quick Start (Usage from Hugging Face)

1. Install dependencies:

pip install -r requirements.txt

2. Set up your environment:

   • Log in to Hugging Face: huggingface-cli login
   • Or set your token as an environment variable: export HUGGING_FACE_TOKEN='your_token'

3. Load and use models:

# Replace with your repository ID
REPO_ID = "matthewchung74/stock-algo-nvda-model"

from model_loader import load_model, load_all_models

# Load a single model
model = load_model('AAPL', repo_id=REPO_ID)
if model:
    predictions = model.predict(your_data, return_probabilities=True)
    signal = model.predict_signal(your_data)

# Load all models
models = load_all_models(repo_id=REPO_ID)
for symbol, model in models.items():
    if model:
        predictions = model.predict(data)

Data Format

Models expect 5-minute OHLCV data with columns:

• timestamp: Datetime of the 5-minute bar
• open, high, low, close: Price data
• volume: Trading volume

Model Features

• Prediction Target: Multi-class classification (Long, Short, Hold signals)
• Prediction Horizon: 12 steps (60 minutes ahead)
• Dynamic Risk Management: Stop-loss and take-profit targets are calculated dynamically using the Average True Range (ATR) to adapt to market volatility.

Files in this Repository

For each model (SYMBOL):

• models/symbol_xgboost_model.pkl: Trained XGBoost classifier
• models/symbol_selected_features.pkl: List of selected features
• models/symbol_model_metadata.json: Training configuration and metrics
• models/symbol_scaler.pkl: Feature scaler (if available)

Support Files

• model_loader.py: Model loading utilities (fetches from HF)
• requirements.txt: Python dependencies
• example_external_usage.py: Usage examples

Notes

• Models were trained on historical 5-minute bar data.
• Performance metrics are available in metadata files.
• Retrain periodically with new data for best results.
• Test thoroughly before live trading.

Generated on: 2025-06-25 12:49:37