Devices and outputs

Device selection is centralized in utils/device_utils.py.

Detection

detect_compute_environment() checks:

• whether nvidia-smi exists
• whether the system is macOS
• whether the machine is Apple Silicon
• whether PyTorch MPS is built and available

Preference:

CUDA -> MPS -> CPU

This mainly affects PyTorch models. XGBoost, LightGBM, and CatBoost only enable GPU parameters on CUDA branches.

Output files

AutoML:

outputs/automl_grid_search_results.json

AutoDL:

outputs/autodl_training_results.json

save_results() creates the output directory if needed and writes UTF-8 JSON with indentation.