Source code for lightautoml.tasks.losses.base

"""Base classes for metric and loss functions."""

from functools import partial
from typing import Any
from typing import Callable
from typing import Dict
from typing import Optional
from typing import Tuple
from typing import Union

import numpy as np

from ..common_metric import _valid_str_metric_names
from ..utils import infer_gib


def fw_rmsle(x, y):
    """Function wrapper for rmsle."""
    return np.log1p(x), y


[docs]class MetricFunc: """Wrapper for metric. Args: metric_func: Callable metric function. m: Multiplier for metric value. bw_func: Backward function. """ def __init__(self, metric_func, m, bw_func): self.metric_func = metric_func self.m = m self.bw_func = bw_func def __call__(self, y_true, y_pred, sample_weight=None) -> float: """Calculate metric.""" y_pred = self.bw_func(y_pred) try: val = self.metric_func(y_true, y_pred, sample_weight=sample_weight) except TypeError: val = self.metric_func(y_true, y_pred) return val * self.m
[docs]class Loss: """Loss function with target transformation.""" @staticmethod def _fw_func(target: Any, weights: Any) -> Tuple[Any, Any]: """Forward transformation. Args: target: Ground truth target values. weights: Item weights. Returns: Tuple (target, weights) without transformation. """ return target, weights @staticmethod def _bw_func(pred: Any) -> Any: """Backward transformation for predicted values. Args: pred: Predicted target values. Returns: Pred without transformation. """ return pred @property def fw_func(self): """Forward transformation for target values and item weights. Returns: Callable transformation. """ return self._fw_func @property def bw_func(self): """Backward transformation for predicted values. Returns: Callable transformation. """ return self._bw_func
[docs] def metric_wrapper( self, metric_func: Callable, greater_is_better: Optional[bool], metric_params: Optional[Dict] = None, ) -> Callable: """Customize metric. Args: metric_func: Callable metric. greater_is_better: Whether or not higher value is better. metric_params: Additional metric parameters. Returns: Callable metric. """ if greater_is_better is None: greater_is_better = infer_gib(metric_func) m = 2 * float(greater_is_better) - 1 if metric_params is not None: metric_func = partial(metric_func, **metric_params) return MetricFunc(metric_func, m, self._bw_func)
[docs] def set_callback_metric( self, metric: Union[str, Callable], greater_is_better: Optional[bool] = None, metric_params: Optional[Dict] = None, task_name: Optional[Dict] = None, ): """Callback metric setter. Args: metric: Callback metric greater_is_better: Whether or not higher value is better. metric_params: Additional metric parameters. task_name: Name of task. Note: Value of ``task_name`` should be one of following options: - `'binary'` - `'reg'` - `'multiclass'` """ assert task_name in [ "binary", "reg", "multiclass", ], "Incorrect task name: {}".format(task_name) self.metric = metric if metric_params is None: metric_params = {} if type(metric) is str: metric_dict = _valid_str_metric_names[task_name] self.metric_func = self.metric_wrapper(metric_dict[metric], greater_is_better, metric_params) self.metric_name = metric else: self.metric_func = self.metric_wrapper(metric, greater_is_better, metric_params) self.metric_name = None