sklearn使用StratifiedShuffleSplit分层抽样踩的坑

发现问题

新手入坑机器学习，使用StratifiedShuffleSplit创建测试集时，一直报如下错误：
ValueError: The least populated class in y has only 1 member, which is too few. The minimum number of groups for any class cannot be less than 2.

踩坑代码

import pandas as pd
from sklearn.model_selection import StratifiedShuffleSplit
housing = pd.read_csv(os.path.join('base_path', 'kc_house_data.csv'))

housing["price_grade"] = np.round(housing["price"] / 100000)
# pandas的where在否的时候，会取替换值
housing["price_grade"].where(housing["price_grade"] < 1000, 1000, True)
strat_train_set, strat_test_set = None, None
# 然后使用分层采样
split = StratifiedShuffleSplit(n_splits=1, test_size=0.2, random_state=42)
for train_index, test_index in split.split(housing, housing["price_grade"]):
    strat_train_set = housing.loc[train_index]
    strat_test_set = housing.loc[test_index]

如上，代码和网上教程都差不多，然鹅他一直报上边的错。

解题思路

分析错误：

ValueError: The least populated class in y has only 1 member, which is too few. The minimum number of groups for any class cannot be less than 2.
ValueError:y中填充最少的类只有1个成员，这太少了。任何类别的最小组数不能少于2个

查看price_grade的值分布情况 housing["price_grade"].value_counts()，存在只有一个成员的类别

解决方法

重新调整分组方法，确保每个类别中至少有两个成员。

housing["price_grade"] = np.ceil(housing["price"] / 20000)
housing["price_grade"].where(housing["price_grade"] < 50, 50, True)
housing["price_grade"].value_counts()

调整过后，最小的类别下有三个成员，再次执行就没问题了。其统计信息如下: