sklearn.metrics
一、回归问题
https://www.jianshu.com/p/9ee85fdad150
- MSE, Mean Squared Error), 均方误差,
sklearn.metrics. mean_squared_error - RMSE,Root Mean Squard Error,均方根误差
- MAE, Mean Absolute Error 平均绝对误差
sklearn.metrics. mean_absolute_error - R-Squared,
sklearn.metrics. r2_score
二、分类问题
基础概念:
- True/False(事实的), Positives(正-阳)/Negatives(负-阴)(预测的)
- 混淆矩阵 confusion_matrix
_ | 预测-Yes(P) | 预测-No(N) |
---|---|---|
实际-Yes | TP-真正,预测为正,实际为正 | FN-假负,将正类预测为负类数→漏报 (Type II error) |
实际-No | FP-假正,将负类预测为正类数误报 (Type I error) | TN-真负,预测为正,实际为负 |
指标:
- Accuracy 准确率:(TP+TN)/ALL
accuracy_score - precision, 精确率/精度/查准率 TP/ (TP+FP)
2.recall, 召回率/查全率/真正例率tpr TP/ (TP+FN)
recall_score
3.f1-score, F值/综合评价指标,2TP/(2TP + FP + FN), F1值就是精确值和召回率的调和均值,P和R指标有的时候是矛盾的,综合考虑精确率(precision)和召回率(recall)这两个度量值。很容易理解,F1综合了P和R的结果,当F1较高时则比较说明实验方法比较理想
4.ROC
roc_curve
5.AUC
roc_auc_score - 假正例率:FPR =FP/ (FP+TN)
8、PR(Precision-Recall)曲线
metrics.classification_report
sklearn.metrics中的评估方法介绍(accuracy_score, recall_score, roc_curve, roc_auc_score, confusion_matrix)
https://blog.csdn.net/CherDW/article/details/55813071
三、排序问题
1.auc
- ndcg_score
Normalized Discounted Cumulative Gain(归一化折损累计增益)
NDCG用作排序结果的评价指标,评价排序的准确性。 - MAP(Mean Average Precision)平均精度均值。
1、准确率(Accuracy)
2、错误率(Error rate)
3、灵敏度/特效度(sensitive)
https://blog.csdn.net/quiet_girl/article/details/70830796
https://tracholar.github.io/machine-learning/2018/01/26/auc.html
https://blog.argcv.com/articles/1036.c
https://blog.csdn.net/quiet_girl/article/details/70830796
网友评论