LGBLoss

class lightautoml.tasks.losses.lgb.LGBLoss(loss, loss_params=None, fw_func=None, bw_func=None)[source]

Bases: lightautoml.tasks.losses.base.Loss

Loss used for LightGBM.

__init__(loss, loss_params=None, fw_func=None, bw_func=None)[source]
Parameters
  • loss (Union[str, Callable]) – Objective to optimize.

  • loss_params (Optional[Dict]) – additional loss parameters. Format like in lightautoml.tasks.custom_metrics.

  • fw_func (Optional[Callable]) – forward transformation. Used for transformation of target and item weights.

  • bw_func (Optional[Callable]) – backward transformation. Used for predict values transformation.

Note

Loss can be one of the types:

  • Str: one of default losses (‘auc’, ‘mse’, ‘mae’, ‘logloss’, ‘accuray’, ‘r2’, ‘rmsle’, ‘mape’, ‘quantile’, ‘huber’, ‘fair’) or another lightgbm objective.

  • Callable: custom lightgbm style objective.

metric_wrapper(metric_func, greater_is_better, metric_params=None)[source]

Customize metric.

Parameters
  • metric_func (Callable) – Callable metric.

  • greater_is_better (Optional[bool]) – Whether or not higher value is better.

  • metric_params (Optional[Dict]) – Additional metric parameters.

Return type

Callable

Returns

Callable metric, that returns (‘Opt metric’, value, greater_is_better).

set_callback_metric(metric, greater_is_better=None, metric_params=None, task_name=None)[source]

Callback metric setter.

Parameters

Note

Value of task_name should be one of following options:

  • ‘binary’

  • ‘reg’

  • ‘multiclass’