如何计算每组中给定事件以来的天数

Question

下面是一个示例数据框：

df = pd.DataFrame({'StudentName': ['Anil','Ramu','Ramu','Anil','Peter','Peter','Anil','Ramu','Peter','Anil'],
                   'ExamDate': ['2021-01-10','2021-01-20','2021-02-22','2021-03-30','2021-01-04','2021-06-06','2021-04-30','2021-07-30','2021-07-08','2021-09-07'],
                   'Result': ['Fail','Pass','Fail','Pass','Pass','Pass','Pass','Pass','Fail','Pass']})

  StudentName    ExamDate Result
0        Anil  2021-01-10   Fail
1        Ramu  2021-01-20   Pass
2        Ramu  2021-02-22   Fail
3        Anil  2021-03-30   Pass
4       Peter  2021-01-04   Pass
5       Peter  2021-06-06   Pass
6        Anil  2021-04-30   Pass
7        Ramu  2021-07-30   Pass
8       Peter  2021-07-08   Fail
9        Anil  2021-09-07   Pass

对于每一行，我想计算自该学生上次考试失败以来的天数：

df = pd.DataFrame({'StudentName': ['Anil','Ramu','Ramu','Anil','Peter','Peter','Anil','Ramu','Peter','Anil'],
                   'ExamDate': ['2021-01-10','2021-01-20','2021-02-22','2021-03-30','2021-01-04','2021-06-06','2021-04-30','2021-07-30','2021-07-08','2021-09-07'],
                   'Result': ['Fail','Pass','Fail','Pass','Pass','Pass','Pass','Pass','Fail','Pass'],
                   'LastFailedDays': [0, 0, 0, 79, 0, 0, 110, 158, 0, 240]})

  StudentName    ExamDate Result  LastFailedDays
0        Anil  2021-01-10   Fail               0
1        Ramu  2021-01-20   Pass               0
2        Ramu  2021-02-22   Fail               0
3        Anil  2021-03-30   Pass              79
4       Peter  2021-01-04   Pass               0
5       Peter  2021-06-06   Pass               0
6        Anil  2021-04-30   Pass             110
7        Ramu  2021-07-30   Pass             158
8       Peter  2021-07-08   Fail               0
9        Anil  2021-09-07   Pass             240

例如：

Anil 在 2021 年 1 月 10 日失败，因此对于该行，它将是零天。
Anil 的下一个成功记录是在 2021-03-30，因此该行的天数将是他上一个失败日期 2021-01-10 到 2021-03- 的天数30，也就是79天。
Anil 的第三次记录也是成功的，是在 2021-04-30，所以还有天数，从 2021-01-10（他最后一次失败的日期）到 2021- 的天数04-30，也就是110天

常规循环是可行的，但我正在寻找更传统的 Pandas 解决方案。我猜 groupby.

是可能的

Answer 1

我终于想出了一个可行的解决方案。

# Process the data a bit
df['Tmp_Result'] = df['Result'].map({'Pass': 1, 'Fail': 0})
df['ExamDate'] = pd.to_datetime(df['ExamDate'])

# Create a mask that will be used to group the rows by StudentName + consecutive passed tests after a failed test (including the failed test)
sorted_df = df.sort_values(['StudentName', 'ExamDate']) 
mask = sorted_df.groupby('StudentName')['Tmp_Result'].diff().ne(0).cumsum()
mask[(sorted_df['Tmp_Result'].eq(0) & ~(pd.isna(sorted_df.groupby('StudentName')['Tmp_Result'].shift(-1))))] += 1

df['LastFailedDays'] = df.groupby(mask)['ExamDate'].diff().fillna(pd.Timedelta(0))
df['LastFailedDays'] = df.groupby(mask)['LastFailedDays'].cumsum()

# Cleanup
df = df.drop('Tmp_Result', axis=1)

输出：

>>> df
  StudentName   ExamDate Result LastFailedDays
0        Anil 2021-01-10   Fail         0 days
1        Ramu 2021-01-20   Pass         0 days
2        Ramu 2021-02-22   Fail         0 days
3        Anil 2021-03-30   Pass        79 days
4       Peter 2021-01-04   Pass         0 days
5       Peter 2021-06-06   Pass       153 days
6        Anil 2021-04-30   Pass       110 days
7        Ramu 2021-07-30   Pass       158 days
8       Peter 2021-07-08   Fail         0 days
9        Anil 2021-09-07   Pass       240 days

>>> df.sort_values(['StudentName', 'ExamDate'])
  StudentName   ExamDate Result LastFailedDays
0        Anil 2021-01-10   Fail         0 days
3        Anil 2021-03-30   Pass        79 days
6        Anil 2021-04-30   Pass       110 days
9        Anil 2021-09-07   Pass       240 days
4       Peter 2021-01-04   Pass         0 days
5       Peter 2021-06-06   Pass       153 days
8       Peter 2021-07-08   Fail         0 days
1        Ramu 2021-01-20   Pass         0 days
2        Ramu 2021-02-22   Fail         0 days
7        Ramu 2021-07-30   Pass       158 days

看起来有点可怕，但因为它是矢量化的，所以它应该比任何使用循环的解决方案都快得多。

Answer 2

TL;DR

使用Series.where and groupby.ffill生成每个学生的最后一次失败日期并从ExamDate中减去得到LastFailedDays:

df['ExamDate'] = pd.to_datetime(df['ExamDate'])

df['LastFailedDays'] = (df['ExamDate'].sub(
    df['ExamDate'].where(df['Result'] == 'Fail').groupby(df['StudentName']).ffill()
).dt.days.fillna(0))

#   StudentName    ExamDate  Result  LastFailedDays
# 0        Anil  2021-01-10    Fail             0.0
# 1        Ramu  2021-01-20    Pass             0.0
# 2        Ramu  2021-02-22    Fail             0.0
# 3        Anil  2021-03-30    Pass            79.0
# 4       Peter  2021-01-04    Pass             0.0
# 5       Peter  2021-06-06    Pass             0.0
# 6        Anil  2021-04-30    Pass           110.0
# 7        Ramu  2021-07-30    Pass           158.0
# 8       Peter  2021-07-08    Fail             0.0
# 9        Anil  2021-09-07    Pass           240.0

回复：评论，按多列分组，例如StudentClass 和 StudentName，使用列表作为石斑鱼：

...groupby([df['StudentClass'], df['StudentName']]).ffill()

详情

转换to_datetime:

df['ExamDate'] = pd.to_datetime(df['ExamDate'])

使用Series.where生成每个学生的最后失败日期（这里我将其设为一列以便于可视化）：

df['LastFailedDate'] = df['ExamDate'].where(df['Result'] == 'Fail')

#   StudentName    ExamDate  Result  LastFailedDate
# 0        Anil  2021-01-10    Fail      2021-01-10
# 1        Ramu  2021-01-20    Pass             NaT
# 2        Ramu  2021-02-22    Fail      2021-02-22
# 3        Anil  2021-03-30    Pass             NaT
# 4       Peter  2021-01-04    Pass             NaT
# 5       Peter  2021-06-06    Pass             NaT
# 6        Anil  2021-04-30    Pass             NaT
# 7        Ramu  2021-07-30    Pass             NaT
# 8       Peter  2021-07-08    Fail      2021-07-08
# 9        Anil  2021-09-07    Pass             NaT

使用 groupby.ffill 向前填写每个学生的最后一次失败日期（NaT 如果之前没有失败的考试）：

df['LastFailedDate'] = df['LastFailedDate'].groupby(df['StudentName']).ffill()

#   StudentName    ExamDate  Result  LastFailedDate
# 0        Anil  2021-01-10    Fail      2021-01-10
# 1        Ramu  2021-01-20    Pass             NaT
# 2        Ramu  2021-02-22    Fail      2021-02-22
# 3        Anil  2021-03-30    Pass      2021-01-10
# 4       Peter  2021-01-04    Pass             NaT
# 5       Peter  2021-06-06    Pass             NaT
# 6        Anil  2021-04-30    Pass      2021-01-10
# 7        Ramu  2021-07-30    Pass      2021-02-22
# 8       Peter  2021-07-08    Fail      2021-07-08
# 9        Anil  2021-09-07    Pass      2021-01-10

最后用最后一次失败的日期减去考试日期并使用dt.days提取天数：

df['LastFailedDays'] = df['ExamDate'].sub(df['LastFailedDate']).dt.days.fillna(0)

#   StudentName    ExamDate  Result  LastFailedDate  LastFailedDays
# 0        Anil  2021-01-10    Fail      2021-01-10             0.0
# 1        Ramu  2021-01-20    Pass             NaT             0.0
# 2        Ramu  2021-02-22    Fail      2021-02-22             0.0
# 3        Anil  2021-03-30    Pass      2021-01-10            79.0
# 4       Peter  2021-01-04    Pass             NaT             0.0
# 5       Peter  2021-06-06    Pass             NaT             0.0
# 6        Anil  2021-04-30    Pass      2021-01-10           110.0
# 7        Ramu  2021-07-30    Pass      2021-02-22           158.0
# 8       Peter  2021-07-08    Fail      2021-07-08             0.0
# 9        Anil  2021-09-07    Pass      2021-01-10           240.0

如何计算每组中给定事件以来的天数

How to calculate the number of days since a given event in each group

python

pandas

pandas-groupby

TL;DR

详情