以两个及更多后续 pandas 行为条件（不仅仅是分组计算）

Question

我有 df，其中包含学生姓名、his/her 分数、class 职称和考试日期。我需要添加一个如图所示的列，表示学生的成绩是否提高（3-4 个条件标记，如“分数增加”、“分数减少”、“相等”或“初始成绩”）。我已经根据这个对 df 进行了排序，现在需要比较行和下一个中的一些条件，如果全部为真，则应该 return 一个标记。有没有一种有效的方法来做到这一点（我的实际 table 将包含 1m 行，这就是为什么它不应该消耗内存）？提前谢谢你？

df=pd.DataFrame({"score":[10,20,15,10,20,30],
                   "student":['John', 'Alex', "John", "John", "Alex", "John"],
                   "class":['english', 'math', "english",'math','math', 'english'],
                 "date":['01/01/2022','02/01/2022', '05/01/2022', '17/02/2022', '02/01/2022', '03/01/2022']})

df=df.sort_values(['student','class', 'date'])

Answer 1

使用 groupby 和 diff() 获取分数变化，然后使用 numpy.select:

分配值

import numpy as np

changes = df.groupby(["student","class"], sort=False)["score"].diff()
df["progress"] = np.select([changes.eq(0),changes.gt(0),changes.lt(0)],
                           ["equal score","score increased","score decreased"], 
                           "initial")

>>> df
   score student    class        date         progress
1     20    Alex     math  02/01/2022          initial
4     20    Alex     math  02/01/2022      equal score
0     10    John  english  01/01/2022          initial
5     30    John  english  03/01/2022  score increased
2     15    John  english  05/01/2022  score decreased
3     10    John     math  17/02/2022          initial

Answer 2

您可以使用 groupby.diff to compute the difference, then numpy.sign to get the sign and map the texts you want. Use fillna 作为默认值（“初始”）：

df['progress'] = (np.sign(df.groupby(['student', 'class'])
                            ['score'].diff())
                    .map({0: 'equal', 1: 'increases', -1: 'decreases'})
                    .fillna('initial')
                  )

输出：

   score student    class        date   progress
1     20    Alex     math  02/01/2022    initial
4     20    Alex     math  02/01/2022      equal
0     10    John  english  01/01/2022    initial
5     30    John  english  03/01/2022  increases
2     15    John  english  05/01/2022  decreases
3     10    John     math  17/02/2022    initial

Answer 3

这是我使用的渐进式方法

df['RN'] = df.sort_values(['date'], ascending=[True]).groupby(['student', 'class']).cumcount() + 1
#df.sort_values(['student', 'RN']) #To visually see progress of student before changes
df['Progress'] = df['RN'].apply(lambda x : str(x).replace('1', 'initial'))
df = df.sort_values(['student', 'RN'])
df['score_shift'] = df['score'].shift()
df['score_shift'].fillna(0, inplace = True)
df['score_shift'] = df['score_shift'].astype(int)
condlist = [df['Progress'] == 'initial', df['score_shift'] == df['score'], df['score_shift'] > df['score'], df['score_shift'] < df['score']]
choicelist = ['initial', 'equal', 'decrease', 'increase']
df['Progress'] = np.select(condlist, choicelist)
df

以两个及更多后续 pandas 行为条件（不仅仅是分组计算）

Condition on two and more subsequent pandas rows (not just grouped calculations)

python

row

conditional-statements

pandas