机器学习 分类问题评价指标

一·交叉验证

二·混淆矩阵

TP 正正  FP 负正   FN 正负  TN负负

from sklearn.model_selection import cross_val_predict

y_train_pred = cross_val_predict(sgd_clf,X_train,y_train_5,cv=3) #获取预测值 

from sklearn.model_metrics import confusion_matrix
confusion_matrix(y_train_5,y_train_pred)    #获取混淆矩阵 

---

三·精确率，召回率

from sklearn.metrics import precision_core, recall_score
precision_scorey=(Y_train_5,y_train_pred)

recall_score(y_train_5,y_train-pred)  #获取精确率召回率

四·F1指标

from sklearn.metrics import fl_score
fl_score(y_train_5,y_train_pred)   #获取F1指标

五·分类阈值对结果的影响

y_score =sgd_clf.decision_function([X[35000]])  #获取当前概率阈值
y_scoresy

t= 50000 #设置新阈值
y_pred=(y_scores >t)

阈值越低，召回率越高，精度越低

反之亦然

六·ROC曲线

from sklearn.metrics import precision_recall_curve
precisions, recalls, thresholds = precision_recall_curve(y_train_5, y_scores)  #获取这三个值

七·ROC curves 

from sklearn.metrics import roc_auc_score
roc_auc_score(y_train_5,y_scores)   #获取AUC的值