stouputils.data_science.scripts.routine module#

routine(
default_input: str = '/home/runner/work/stouputils/stouputils/data/aug_hip_implant_preprocessed',
default_based_of: str = 'auto',
default_transfer_learning: str = 'imagenet',
default_grouping_strategy: str = 'none',
default_kfold: int = 0,
default_verbose: int = 100,
loading_type: Literal['image'] = 'image',
grid_search_param_grid: dict[str, list[Any]] | None = None,
add_to_train_only: list[str] | None = None,
) → None[source]#

Main function of the script for training and evaluating machine learning models.

This function handles the entire workflow for model training and evaluation, including: - Parsing command-line arguments (default values are set in the function signature) - Loading and preparing datasets with configurable grouping strategies - Supporting transfer learning from various sources - Enabling K-fold cross-validation, LeavePOut or LeaveOneOut - Providing grid search capabilities for hyperparameter optimization - Incorporating additional training data from specified paths

Parameters:

  • default_input (str) – Default path to the dataset to use.

  • default_based_of (str) – Default path to the base dataset for filtering train/test data.

  • default_transfer_learning (str) – Default transfer learning source.

  • default_grouping_strategy (str) – Default grouping strategy for the dataset.

  • default_kfold (int) – Default number of folds for k-fold cross validation.

  • default_verbose (int) – Default verbosity level.

  • loading_type (Literal["image"]) – Type of data to load, currently only supports “image”.

  • grid_search_param_grid (dict[str, list[Any]] | None) – Parameters grid for hyperparameter optimization.

  • add_to_train_only (list[str] | None) – List of paths to additional training datasets.

Returns:

This function does not return anything.

Return type:

None