environment.yml

name: foxml_env
channels:
  - conda-forge
  - defaults
  - pytorch
  - nvidia

dependencies:
  # Python
  - python=3.10

  # Core Scientific Computing
  - numpy>=2.0.0,<3.0.0  # Tested with 2.2.6
  - pandas>=2.0.0,<3.0.0  # Tested with 2.3.3
  - scipy>=1.10.0,<2.0.0  # Tested with 1.15.3
  
  # Data Handling & Storage
  - pyarrow>=20.0.0  # Tested with 22.0.0
  - fastparquet>=2024.0.0  # Tested with 2024.11.0
  - h5py>=3.15.0  # Tested with 3.15.1
  - tables>=3.8.0
  
  # Machine Learning Core
  - scikit-learn>=1.7.0  # Tested with 1.7.2
  - joblib>=1.5.0  # Tested with 1.5.2
  
  # Deep Learning (Optional - if using neural networks)
  - pytorch>=2.0.0
  - torchvision>=0.15.0
  - pytorch-cuda=12.1  # Match your CUDA version (13.0 compatible)
  
  # TensorFlow (with CUDA support)
  - tensorflow>=2.19.0  # Tested with 2.19.1 (CUDA 12.9)
  - keras>=3.12.0  # Tested with 3.12.0
  
  # XGBoost (must be built from source with CUDA for GPU support)
  # See SCRIPTS/build_xgboost_cuda.sh
  # Tested with 3.2.0-dev (built from source)
  
  # Statistical & Time Series
  - statsmodels>=0.14.0
  - ta-lib>=0.4.0  # Technical analysis
  
  # Visualization
  - matplotlib>=3.10.0  # Tested with 3.10.7
  - seaborn>=0.12.0
  - plotly>=6.0.0  # Tested with 6.4.0
  
  # Configuration & Utilities
  - pyyaml>=6.0
  - toml>=0.10.0
  - python-dotenv>=1.0.0
  
  # Progress Bars & Logging
  - tqdm>=4.65.0
  - colorlog>=6.7.0
  
  # System Utilities
  - psutil>=7.0.0  # Tested with 7.0.0
  
  # Financial Data
  # NOTE: Market data libraries are NOT included. Production customers are expected to plug in their own data feeds.
  # FoxML Core is a pipeline infrastructure, not a data provider.
  
  # API & Networking
  - requests>=2.31.0
  - aiohttp>=3.8.0
  - websockets>=11.0
  
  # Database (if needed)
  - sqlalchemy>=2.0.0
  - psycopg2>=2.9.0
  
  # Jupyter & Development Tools
  - jupyter>=1.0.0
  - jupyterlab>=4.0.0
  - ipython>=8.12.0
  - ipywidgets>=8.0.0
  
  # Testing
  - pytest>=7.3.0
  - pytest-cov>=4.1.0
  
  # Code Quality
  - black>=23.0.0
  - ruff>=0.0.270
  - mypy>=1.3.0
  
  # Build Tools (for LightGBM CUDA build)
  - cmake>=3.26.0
  - make
  - gcc_linux-64>=11.0.0
  - gxx_linux-64>=11.0.0
  
  # CUDA Toolkit (matches system CUDA 12.9)
  - cuda-toolkit=12.9  # Tested with CUDA 12.9
  - cuda-nvcc=12.9  # For building XGBoost/LightGBM with CUDA
  - cudnn>=8.9.0
  
  # Pip packages (installed via pip after conda)
  - pip>=23.0.0
  
  # Pip-only packages
  - pip:
    # Broker and execution adapters are NOT included in this environment.
    # FoxML Core is broker-agnostic; production execution is handled via internal/commercial adapters.
    # Trading integration modules have been removed from the core repository.
    
    # Additional ML Libraries
    - ngboost>=0.4.0
    - optuna>=3.2.0  # Hyperparameter optimization
    - shap>=0.42.0  # Model explainability
    
    # Time Series Specific
    - arch>=6.2.0  # ARCH/GARCH models
    
    # Monitoring & Alerts
    - ntfy>=2.7.0  # Notifications
    
    # Additional utilities
    - python-dateutil>=2.8.2
    - pytz>=2023.3
    
    # Polars (for efficient cross-sectional data processing)
    - polars>=1.35.0  # Tested with 1.35.2
    
    # NOTE: LightGBM with CUDA must be built from source
    # Do NOT install lightgbm via pip - it's CPU only
    # After creating this environment, build LightGBM with:
    #   bash SCRIPTS/build_lightgbm_cuda.sh
    #
    # NOTE: XGBoost with CUDA must be built from source
    # Do NOT install xgboost via pip/conda - GPU support not available
    # After creating this environment, build XGBoost with:
    #   bash SCRIPTS/build_xgboost_cuda.sh

# Post-installation instructions:
# 1. Create environment:
#    conda env create -f environment.yml
#
# 2. Activate environment:
#    conda activate foxml_env
#
# 3. Set MKL threading layer to avoid OpenMP conflicts:
#    export MKL_THREADING_LAYER=GNU
#    (This prevents Intel OpenMP conflicts with LightGBM/XGBoost)
#    (Or add to ~/.bashrc/.zshrc for persistence)
#
# 4. Build LightGBM with CUDA (RTX 3080):
#    bash SCRIPTS/build_lightgbm_cuda.sh
#
# 5. Build XGBoost with CUDA (RTX 3080):
#    bash SCRIPTS/build_xgboost_cuda.sh
#
# 6. Verify GPU setup:
#    python SCRIPTS/check_gpu_setup.py
#
# 7. Test feature selection:
#    python SCRIPTS/select_features.py --symbols AAPL