用“其他”python重命名不太频繁的类别

Question

在我的数据框中，我有一些包含 100 多个不同类别的分类列。我想按最常见的类别对类别进行排名。我保留前 9 个最常见的类别，而不太常见的类别会自动将它们重命名为：OTHER

例子：

这里是我的 df：

print(df)

    Employee_number                 Jobrol
0                 1        Sales Executive
1                 2     Research Scientist
2                 3  Laboratory Technician
3                 4        Sales Executive
4                 5     Research Scientist
5                 6  Laboratory Technician
6                 7        Sales Executive
7                 8     Research Scientist
8                 9  Laboratory Technician
9                10        Sales Executive
10               11     Research Scientist
11               12  Laboratory Technician
12               13        Sales Executive
13               14     Research Scientist
14               15  Laboratory Technician
15               16        Sales Executive
16               17     Research Scientist
17               18     Research Scientist
18               19                Manager
19               20        Human Resources
20               21        Sales Executive


valCount = df['Jobrol'].value_counts()

valCount

Sales Executive          7
Research Scientist       7
Laboratory Technician    5
Manager                  1
Human Resources          1

我保留前 3 个类别，然后将其余类别重命名为“OTHER”，我应该如何继续？

谢谢。

Answer 1

一线解决方案：

limit = 500
df['Jobrol'] = df['Jobrol'].map({x[0]: x[0] if x[1] > limit else 'other' for x in dict(df['Jobrol'].value_counts()).items()})

Answer 2

将您的系列转换为分类，提取计数不在前 3 的类别，添加一个新类别，例如'Other'，然后替换之前计算的类别：

df['Jobrol'] = df['Jobrol'].astype('category')

others = df['Jobrol'].value_counts().index[3:]
label = 'Other'

df['Jobrol'] = df['Jobrol'].cat.add_categories([label])
df['Jobrol'] = df['Jobrol'].replace(others, label)

注意：通过df['Jobrol'].cat.rename_categories(dict.fromkeys(others, label)) 重命名类别来组合类别是诱人的，但这不起作用，因为这将意味着多个相同标记的类别，这是'不可能。

---

上面的解决方案可以适应count过滤。例如，要仅包含计数为 1 的类别，您可以这样定义 others：

counts = df['Jobrol'].value_counts()
others = counts[counts == 1].index

Answer 3

将value_counts 与numpy.where 一起使用：

need = df['Jobrol'].value_counts().index[:3]
df['Jobrol'] = np.where(df['Jobrol'].isin(need), df['Jobrol'], 'OTHER')

valCount = df['Jobrol'].value_counts()
print (valCount)
Research Scientist       7
Sales Executive          7
Laboratory Technician    5
OTHER                    2
Name: Jobrol, dtype: int64

另一种解决方案：

N = 3
s = df['Jobrol'].value_counts()
valCount = s.iloc[:N].append(pd.Series(s.iloc[N:].sum(), index=['OTHER']))
print (valCount)
Research Scientist       7
Sales Executive          7
Laboratory Technician    5
OTHER                    2
dtype: int64