数据样本不平衡时处理方法（imblearn实现，pandas/numpy转换）

def getDataAndSubsample(sample_strategy="under"):
    """
    获得数据并采样
    :param sample_strategy: 采样策略
    :return:                Dataframe数据
    """
    # pandas读取
    new_path = '../newData/结果/5#/total/all_20180131_rounding_five.csv'  # 5分类的（0,1,2,3,4）

    # tmp_data  = pd.read_csv(new_path, index_col=[0])		# csv文件如果第一列有索引就用这个
    tmp_data = pd.read_csv(new_path)						# csv文件如果第一列没索引就用这个

    under_map_dict = {0: 300, 1: 300, 2: 300, 3: 300, 4: 300}		# 设置欠采样各类数量
    over_map_dict = {0: 4000, 1: 4000, 2: 4000, 3: 4000, 4: 4000}	# 设置过采样各类数量
	
	# 选择下采样
    if sample_strategy == "under":
        from imblearn.under_sampling import RandomUnderSampler
        # 通过设置RandomUnderSampler中的replacement=True参数, 可以实现自助法(boostrap)抽样
        # （仅在二分类时可用）通过设置RandomUnderSampler中的ratio参数（<=1的float类型）,可以设置数据采样比例
        # （多分类时）不能写ratio,只能写sampling_strategy=dict！，像下面一样分别设置5个label个数
        rus = RandomUnderSampler(sampling_strategy=under_map_dict, random_state=0, replacement=True)  # 采用随机欠采样（下采样）
        x_resample, y_resample = rus.fit_sample(np.array(tmp_data.iloc[:, 0:-1]),
                                                (np.array(tmp_data.iloc[:, -1])).astype(int))
	# 选择SMOTETomek过采样
    elif sample_strategy == "over1":
        from imblearn.combine import SMOTETomek                  # 过采样
        smote_tomek = SMOTETomek(sampling_strategy=over_map_dict)     # 取数据
        x_resample, y_resample = smote_tomek.fit_sample(np.array(tmp_data.iloc[:,0:-1]), (np.array(tmp_data.iloc[:, -1])).astype(int))

	# 选择SMOTEENN过采样
    elif sample_strategy == "over2":
        from imblearn.combine import SMOTEENN
        smote_enn = SMOTEENN(random_state=0, sampling_strategy=over_map_dict)
        x_resample, y_resample = smote_enn.fit_sample(np.array(tmp_data.iloc[:,0:-1]), (np.array(tmp_data.iloc[:, -1])).astype(int))

    print(f"len of x_resample={len(x_resample)}")
    print(f"len of y_resample={len(y_resample)}")

	# 矩阵合并，为了使两个矩阵的dimension相同
    total_np_data = np.column_stack((x_resample, y_resample[:,None]))	
	
	# 特征表头
    featureClass_list_columns = featureClass_list.copy()
    featureClass_list_columns.append('score')
    
    # numpy转pandas
    total_pd_data = pd.DataFrame(total_np_data, columns=featureClass_list_columns)

    print(total_pd_data)
    print(total_pd_data.columns)

getDataAndSubsample(sample_strategy="under")

原理过一年再补，急着看的请移步其它博客