src.config

Configuration loader module. Uses Hydra's Compose API to load config.yaml with support for CLI overrides.

Architecture

Unlike the typical @hydra.main decorator approach, this module uses Hydra's Compose API for maximum flexibility:

  • initialize_config_dir() points Hydra at the project root
  • compose() loads config.yaml and applies any overrides
  • OmegaConf.to_container() converts the result to a plain Python dict
  • The config is loaded once at import time and shared globally as CONFIG

This avoids Hydra's side-effects (output directories, working directory changes) while keeping its powerful override system.

Module-Level Constants

PROJECT_ROOT

PROJECT_ROOT: Path

The absolute path to the project root directory (where config.yaml lives). Resolved at import time by walking up from config.py until config.yaml is found.

CONFIG

CONFIG: dict

The parsed configuration dictionary, loaded once at import time via load_config(). If CLI arguments contain Hydra-style overrides (e.g. training.n_trials=5), they are applied automatically.

Example usage:

from src.config import CONFIG, resolve_path

train_csv = resolve_path(CONFIG["data"]["train_csv"])
n_trials = CONFIG["training"]["n_trials"]

Functions

get_project_root

def get_project_root() -> Path

Walks up the directory tree from the location of config.py until it finds a directory containing config.yaml.

Returns:

Type Description
Path Absolute path to the project root directory

Raises:

Exception Condition
FileNotFoundError If config.yaml is not found in any parent directory

_parse_cli_overrides

def _parse_cli_overrides() -> list[str]

Extracts Hydra-style key=value overrides from sys.argv. This is necessary because the Compose API (unlike @hydra.main) does not automatically parse CLI arguments.

Filtering logic:

  • Includes arguments that contain =
  • Excludes arguments starting with - (flags like --verbose)

Returns:

Type Description
list[str] List of override strings, e.g. ["training.n_trials=5", "training.cv_folds=3"]

Example:

# Running:
python -m src.train training.n_trials=5 --verbose

# _parse_cli_overrides() returns:
# ["training.n_trials=5"]
# (--verbose is excluded because it starts with "-")

load_config

def load_config(overrides: list[str] | None = None) -> dict

Loads config.yaml via Hydra's Compose API with optional overrides.

Parameters:

Parameter Type Default Description
overrides list[str] \| None None Hydra override strings. If None, auto-detects from CLI args via _parse_cli_overrides().

Returns:

Type Description
dict Parsed YAML configuration as a nested dictionary with all overrides applied

Example — programmatic overrides:

from src.config import load_config

# Override programmatically (ignores CLI args)
cfg = load_config(overrides=["training.n_trials=10", "preprocessing.test_size=0.3"])
print(cfg["training"]["n_trials"])  # 10

Example — automatic CLI detection:

# When called with no arguments, reads overrides from sys.argv
cfg = load_config()  # picks up CLI args automatically

resolve_path

def resolve_path(relative_path: str) -> Path

Converts a config-relative path string to an absolute path by prepending PROJECT_ROOT.

Parameters:

Parameter Type Description
relative_path str A path relative to the project root (e.g. "data/raw/train.csv")

Returns:

Type Description
Path Absolute path

Example:

from src.config import resolve_path

path = resolve_path("models/best_model.pkl")
# → /absolute/path/to/mlops-lab0/models/best_model.pkl