如何修改此代码以实现 Smote 过采样和交叉验证管道以解决多类分类问题？

Question

imba_pipeline = make_pipeline(SMOTE(random_state=42), 
                              DecisionTreeClassifier())
cross_val_score(imba_pipeline, X_train1, y_train1, scoring='recall', cv=kf)

当我运行它时，它给出了以下错误：

ValueError：目标是多类但平均值='二进制'。请选择另一个平均设置，[None, 'micro', 'macro', 'weighted'] 之一。

    Out[102]:
    array([nan, nan, nan, nan, nan])

Answer 1

如果您使用 score='recall' ，它会调用 sklearn.metrics.recall_score 并使用默认值 average = "binary" 。一种方法是创建自定义评分：

from sklearn.datasets import load_iris
from sklearn.tree import DecisionTreeClassifier
from imblearn.over_sampling import SMOTE
from imblearn.pipeline import make_pipeline
from sklearn.model_selection import cross_val_score
from sklearn.metrics import recall_score
from sklearn.metrics import make_scorer

X,y = load_iris(return_X_y=True)

imba_pipeline = make_pipeline(SMOTE(random_state=42), 
                              DecisionTreeClassifier())

multi_recall = make_scorer(recall_score, average="micro")

cross_val_score(imba_pipeline, X, y, scoring=multi_recall, cv=5)

array([0.96666667, 0.96666667, 0.9       , 0.93333333, 1.        ])