# -*- coding:utf-8 -*-
__author__ = 'yangjian'
import numpy as np
from sklearn.metrics import roc_auc_score, mean_squared_log_error, accuracy_score, \
mean_squared_error, mean_absolute_error, r2_score, precision_score, recall_score, f1_score, log_loss
from ..utils import consts
[docs]def calc_score(y_true, y_proba, y_preds, metrics, task, pos_label=1):
score = {}
if len(y_proba.shape) == 2 and y_proba.shape[-1] == 1:
y_proba = y_proba.reshape(-1)
if len(y_preds.shape) == 2 and y_preds.shape[-1] == 1:
y_preds = y_preds.reshape(-1)
for metric in metrics:
if callable(metric):
score[metric.__name__] = metric(y_true, y_preds)
else:
metric = metric.lower()
if task == consts.TASK_MULTICLASS:
average = 'micro'
else:
average = 'binary'
if metric == 'auc':
if len(y_proba.shape) == 2:
# ! When sklearn version > 0.22.0 support multi class
score['auc'] = roc_auc_score(y_true, y_proba, multi_class='ovo')
else:
score['auc'] = roc_auc_score(y_true, y_proba)
elif metric == 'accuracy':
if y_preds is None:
score['accuracy'] = 0
else:
score['accuracy'] = accuracy_score(y_true, y_preds)
elif metric == 'log_loss':
score['log_loss'] = log_loss(y_true, y_proba)
elif metric == 'recall':
score['recall'] = recall_score(y_true, y_preds, average=average, pos_label=pos_label)
elif metric == 'precision':
score['precision'] = precision_score(y_true, y_preds, average=average, pos_label=pos_label)
elif metric == 'f1':
score['f1'] = f1_score(y_true, y_preds, average=average, pos_label=pos_label)
elif metric == 'mse':
score['mse'] = mean_squared_error(y_true, y_preds)
elif metric == 'mae':
score['mae'] = mean_absolute_error(y_true, y_preds)
elif metric == 'msle':
score['msle'] = mean_squared_log_error(y_true, y_preds)
elif metric == 'rmse':
score['rmse'] = np.sqrt(mean_squared_error(y_true, y_preds))
elif metric == 'rootmeansquarederror':
score['rootmeansquarederror'] = np.sqrt(mean_squared_error(y_true, y_preds))
elif metric == 'r2':
score['r2'] = r2_score(y_true, y_preds)
return score