基于平均值聚类DataFrame相似值时遇TypeError的排查与修复

嘿，我一眼就看出问题出在哪了！你遇到的TypeError: cannot perform reduce with flexible type错误，根源完全不是Consumption列有字符串，而是你在计算均值的时候传错了参数！

错误原因分析 #

你写的np.mean(['Consumption'])是在对字符串列表['Consumption']求均值——这显然是南辕北辙！numpy没法对字符串做数值运算，自然会抛出“无法对灵活类型执行归约操作”的错误。你真正需要的是整个DataFrame中Consumption列的数值均值，而不是这个字符串的“均值”。

修复方案一：修改自定义函数 #

我们先提前计算好全局的Consumption均值，再在聚类函数里用这个预计算的值来判断，代码如下：

import numpy as np
import pandas as pd

# 构造你的数据
data = {
    'Zone': ['North', 'South', 'East', 'North', 'North2', 'South'],
    'Consumption': [1, 3, 10, 8, 0, 5]
}
df = pd.DataFrame(data)

# 预先计算Consumption列的数值均值（这才是正确的方式！）
mean_consumption = df['Consumption'].mean()

def Clustering(row):
    if row['Consumption'] < 0.5 * mean_consumption:
        val = 'E'
    elif row['Consumption'] < 0.75 * mean_consumption:
        val = 'D'
    elif row['Consumption'] < 1 * mean_consumption:
        val = 'C'
    elif row['Consumption'] < 1.5 * mean_consumption:
        val = 'B'
    elif row['Consumption'] < 2.5 * mean_consumption:
        val = 'A'
    else:
        val = 'Z'
    return val

# 给DataFrame添加聚类结果列
df['Cluster'] = df.apply(Clustering, axis=1)

修复方案二：用更高效的pd.cut分箱 #

其实pandas自带的pd.cut方法更适合这种基于阈值的分箱聚类，不用写自定义函数，代码更简洁高效：

import numpy as np
import pandas as pd

# 构造数据
data = {
    'Zone': ['North', 'South', 'East', 'North', 'North2', 'South'],
    'Consumption': [1, 3, 10, 8, 0, 5]
}
df = pd.DataFrame(data)

mean_consumption = df['Consumption'].mean()

# 定义分箱边界和对应标签
bins = [-np.inf, 0.5*mean_consumption, 0.75*mean_consumption, mean_consumption, 1.5*mean_consumption, 2.5*mean_consumption, np.inf]
labels = ['E', 'D', 'C', 'B', 'A', 'Z']

# 直接生成聚类结果
df['Cluster'] = pd.cut(df['Consumption'], bins=bins, labels=labels)

运行这两种方案都能得到正确的聚类结果，而且不会再报错啦！

内容的提问来源于stack exchange，提问作者Karthik Apadodharanan