scanpex.ml.lightgbm_args package

Module contents

scanpex.ml.lightgbm_args.multiclass_args(num_class, objective='multiclass', metric='multi_logloss', verbosity=-1, deterministic=True, random_seed=0, num_boost_round=100, force_col_wise=True)[source]

Generate a hyperparameter dictionary for LightGBM multi-class classification.

This helper function constructs the configuration dictionary required to train a LightGBM model. It sets standard defaults for reproducibility and logging.

Parameters:

  • num_class (int) – The number of target classes. This is a required parameter for multi-class objectives.

  • objective (str, optional) – The learning objective. Common values include “multiclass” and “multiclassova”. By default “multiclass”.

  • metric (str, optional) – The metric to be evaluated on the evaluation set. Common values include “multi_logloss” or “multi_error”. By default “multi_logloss”.

  • verbosity (int, optional) – Controls the level of LightGBM logging. < 0 for fatal only, 0 for error, 1 for info. By default -1 (silent).

  • deterministic (bool, optional) – If True, ensures reproducible results. By default True.

  • random_seed (int, optional) – The seed for the random number generator. By default 0.

  • num_boost_round (int, optional) – The number of boosting iterations (trees) to build. By default 100.

  • force_col_wise (bool, optional) – If True, forces column-wise histogram building, which can reduce memory usage and is generally faster on CPUs with many cores. By default True.

Returns:

A dictionary containing the parameter keys and values ready to be passed to LightGBM training functions.

Return type:

dict

scanpex.ml.lightgbm_args.regression_args(objective='regression', metric='l2', verbosity=-1, deterministic=True, random_seed=0, num_boost_round=100, force_col_wise=True)[source]

Generate a hyperparameter dictionary for LightGBM regression tasks.

Parameters:

  • objective (str, optional) – The learning objective. Common values include “regression” (L2), “regression_l1” (L1), or “huber”. By default “regression”.

  • metric (str, optional) – The metric to be evaluated on the evaluation set. Common values include “l2” (MSE), “l1” (MAE), or “rmse”. By default “l2”.

  • verbosity (int, optional) – Controls the level of LightGBM logging. By default -1 (silent).

  • deterministic (bool, optional) – If True, ensures reproducible results. By default True.

  • random_seed (int, optional) – The seed for the random number generator. By default 0.

  • num_boost_round (int, optional) – The number of boosting iterations (trees) to build. By default 100.

  • force_col_wise (bool, optional) – If True, forces column-wise histogram building. By default True.

Returns:

A dictionary containing the parameter keys and values ready to be passed to LightGBM training functions.

Return type:

dict