machine-learning - RandomOverSampler 条件创建均等分布

Question

我目前正在从事一个基于 ML 的项目，我的数据存在轻微的不平衡，需要过度采样技术。特征 (X_train) 维度是 (90664, 190)，目标 (Y_binary_train_trans) 是 (90664, )。但是，代码运行并仍然输出相同的、不均等的目标分布。这是用于 RandomOverSampler 的代码，它也已尝试使用 smote；

counter= Counter(Y_binary_train_trans)
ros= RandomOverSampler(random_state=42)
X_train, Y_binary_train_trans = ros.fit_resample(X_train,Y_binary_train_trans)
counter = Counter(Y_binary_test_trans)

标签： machine-learningscikit-learnimblearn

Answer 1

counter= Counter(Y_binary_train_trans)
ros= RandomOverSampler(random_state=42)
X_train, Y_binary_train_trans = ros.fit_resample(X_train,Y_binary_train_trans)
counter = Counter(Y_binary_test_trans)

至于此代码，您的第二个计数器计算的是测试样本，而不是您实际更改的训练样本！

相反，它应该是：

counter= Counter(Y_binary_train_trans)
ros= RandomOverSampler(random_state=42)
X_train, Y_binary_train_trans = ros.fit_resample(X_train,Y_binary_train_trans)
counter = Counter(Y_binary_train_trans)