imblearn 解决样本不均衡问题，上采样与下采样

from imblearn.over_sampling import RandomOverSampler
from imblearn.under_sampling import RandomUnderSampler

# 上采样
def over_sampling(train_img, train_label):
    rus = RandomOverSampler(random_state=random_seed)
    nsamples, nx, ny, nz, = train_img.shape
    train_img_flatten = train_img.reshape(nsamples, nx * ny * nz)
    X_resampled, y_resampled = rus.fit_resample(train_img_flatten, train_label)
    X_resample = X_resampled.reshape(X_resampled.shape[0], nx, ny, nz)
    return X_resample, y_resampled

解释，简单的对欠采样的少数样本进行随机选择添加。

类似于sklearn的先对类实例化，拿到一个转换器对象，然后调用对象的fit_resample方法，返回值即为上采样后的样本与标签，下采样过程相同。

这里还对一批图像数据进行展平操作，下采样过程相同。

# 下采样
def under_sampling(train_img, train_label):
    rus = RandomUnderSampler(random_state=random_seed, replacement=True)
    n_samples, nx, ny, nz = train_img.shape
    train_img_flatten = train_img.reshape(n_samples, nx * ny * nz)
    x_resampled, y_resampled = rus.fit_resample(train_img_flatten, train_label)
    x_resampled = x_resampled.reshape(x_resampled.shape[0], nx, ny, nz)
    return x_resampled, y_resampled

解释，简单的对多数样本进行随机选择删除

数据增强仅仅用于训练数据集，不需要改变交叉验证集和测试集