Python 熊猫-如何使用更新的事件重新计算预计到达时间

我有一个包含以下事件的数据框：

                       ID           m1          m2          m3   m4
1   xxxx/xxxxx.0183683234   2019-10-28  2019-11-28  2019-11-30  NaT
2   xxxx/xxxxx.0183679721   2019-11-28  2019-11-28         NaT  NaT
4   xxxx/xxxxx.0183888975   2019-11-20  2019-12-10         NaT  NaT

这些事件是按时间顺序发生的。这意味着：

---

m1

for col in df.columns:
    if col != 'ID':
         df[col] = pd.to_datetime(df[col], errors='coerce')

我仍然不知道masterdata是什么样子，但我假设它的行长度与原始数据帧的行长度相同。 这是我的主要数据：

master = pd.DataFrame([['xxxx/xxxxx.0183683234','2 days','9 days'],
                       ['xxxx/xxxxx.0183679721','2 days','6 days'],
                       ['xxxx/xxxxx.0183888975','6 days','1 day']],columns=['ID','M2_M3','M3_M4'])


out = master    
out = out.merge(df, on='ID') #it will the expected output

# variables to new columns
m3_estimated = []
m4_estimated = []

# library to add days in a simple way
from datetime import timedelta  


for li,m2_v in zip(out['M2_M3'].dt.days.astype('int16'),out['m2']):
   if m2_v:
        m3_estimated.append(m2_v + timedelta(days=li))
    else:
        m3_estimated.append(None)


for li,m3_v in zip(out['M3_M4'].dt.days.astype('int16'),out['m3']):
    if m3_v:
        m4_estimated.append(m3_v + timedelta(days=li))
    else:
        m4_estimated.append(None)


out['m3_estimated'] = m3_estimated
out['m4_estimated'] = m4_estimated

print(out)
                      ID   M2_M3   M3_M4  ...  m4 m3_estimated m4_estimated
0  xxxx/xxxxx.0183683234  2 days  9 days  ... NaT   2019-11-30   2019-12-09
1  xxxx/xxxxx.0183679721  2 days  6 days  ... NaT   2019-11-30          NaT
2  xxxx/xxxxx.0183888975  6 days   1 day  ... NaT   2019-12-16          NaT

一种可能的解决办法：

df
            xxxxxxxxxxID         m1         m2         m3  m4  M2_M3  M3_M4
1  xxxx/xxxxx.0183683234 2019-10-28 2019-11-28 2019-11-30 NaT 2 days 9 days
2  xxxx/xxxxx.0183679721 2019-11-28 2019-11-28        NaT NaT 2 days 6 days
4  xxxx/xxxxx.0183888975 2019-11-20 2019-12-10        NaT NaT 6 days 1 days

df.dtypes                                                                                                             
xxxxxxxxxxID             object
m1               datetime64[ns]
m2               datetime64[ns]
m3               datetime64[ns]
m4               datetime64[ns]
M2_M3           timedelta64[ns]
M3_M4           timedelta64[ns]
dtype: object

#This two lines can be put in a timeloop:
df["m3_estimated"]=df.m3.where(~df.m3.isna(), df.m2.add(df.M2_M3))
df["m4_estimated"]=df.m4.where(~df.m4.isna(), df.m3_estimated.add(df.M3_M4))

df
            xxxxxxxxxxID         m1         m2         m3  m4  M2_M3  M3_M4 m3_estimated m4_estimated
1  xxxx/xxxxx.0183683234 2019-10-28 2019-11-28 2019-11-30 NaT 2 days 9 days   2019-11-30   2019-12-09
2  xxxx/xxxxx.0183679721 2019-11-28 2019-11-28        NaT NaT 2 days 6 days   2019-11-30   2019-12-06
4  xxxx/xxxxx.0183888975 2019-11-20 2019-12-10        NaT NaT 6 days 1 days   2019-12-16   2019-12-17

---

df.m4.其中（…）选择m4的值（如果它是固定的），或者使用m3_estimated和m3_m4进行计算。

您的日期时间列始终是datetime吗？您的数据帧中是否有其他日期时间列？您应该将pd.Timedelta列添加到pd.datetime。也就是说，

pd.to_datetime（df['m2']）+pd.to_timedelta（df['m2\u M3']，unit='D'）

如果后者是整数，或者如果不是整数，您应该将其转换为整数。您检查过这个堆栈吗？确实，timedelta函数是用于此的。但我的问题是，当实际情况不是这样时，如何计算估计值null@powerPixie是的，我检查了，但它不是我想要的，我使用它时出现了这个错误AttributeError:只能使用带字符串值的.str访问器，它在pandas中使用np.object\udtype，来自此行：对于li，m2\u v in zip（out['m2\u M3'].str.split（''），out['m2']）：导入numpy作为npIt数据本身肯定有问题。我使用了您示例中可用的数据。“M2\u M3”列中的数据似乎有问题。在不知道冲突原因的情况下是很难帮助的。M2_M3的数据类型是：timedelta64[ns]和空值？我喜欢这个解决方案，它正在工作。。。但是，我必须添加.dt.normalize（）以跳过时间戳值BTW，如何在timeloop中添加行？你这是什么意思？@Haalanam我的意思是定期执行包含数据重读部分和这两行的代码。

master = pd.DataFrame([['xxxx/xxxxx.0183683234','2 days','9 days'],
                       ['xxxx/xxxxx.0183679721','2 days','6 days'],
                       ['xxxx/xxxxx.0183888975','6 days','1 day']],columns=['ID','M2_M3','M3_M4'])


out = master    
out = out.merge(df, on='ID') #it will the expected output

# variables to new columns
m3_estimated = []
m4_estimated = []

# library to add days in a simple way
from datetime import timedelta  


for li,m2_v in zip(out['M2_M3'].dt.days.astype('int16'),out['m2']):
   if m2_v:
        m3_estimated.append(m2_v + timedelta(days=li))
    else:
        m3_estimated.append(None)


for li,m3_v in zip(out['M3_M4'].dt.days.astype('int16'),out['m3']):
    if m3_v:
        m4_estimated.append(m3_v + timedelta(days=li))
    else:
        m4_estimated.append(None)


out['m3_estimated'] = m3_estimated
out['m4_estimated'] = m4_estimated

print(out)
                      ID   M2_M3   M3_M4  ...  m4 m3_estimated m4_estimated
0  xxxx/xxxxx.0183683234  2 days  9 days  ... NaT   2019-11-30   2019-12-09
1  xxxx/xxxxx.0183679721  2 days  6 days  ... NaT   2019-11-30          NaT
2  xxxx/xxxxx.0183888975  6 days   1 day  ... NaT   2019-12-16          NaT

df
            xxxxxxxxxxID         m1         m2         m3  m4  M2_M3  M3_M4
1  xxxx/xxxxx.0183683234 2019-10-28 2019-11-28 2019-11-30 NaT 2 days 9 days
2  xxxx/xxxxx.0183679721 2019-11-28 2019-11-28        NaT NaT 2 days 6 days
4  xxxx/xxxxx.0183888975 2019-11-20 2019-12-10        NaT NaT 6 days 1 days

df.dtypes                                                                                                             
xxxxxxxxxxID             object
m1               datetime64[ns]
m2               datetime64[ns]
m3               datetime64[ns]
m4               datetime64[ns]
M2_M3           timedelta64[ns]
M3_M4           timedelta64[ns]
dtype: object

#This two lines can be put in a timeloop:
df["m3_estimated"]=df.m3.where(~df.m3.isna(), df.m2.add(df.M2_M3))
df["m4_estimated"]=df.m4.where(~df.m4.isna(), df.m3_estimated.add(df.M3_M4))

df
            xxxxxxxxxxID         m1         m2         m3  m4  M2_M3  M3_M4 m3_estimated m4_estimated
1  xxxx/xxxxx.0183683234 2019-10-28 2019-11-28 2019-11-30 NaT 2 days 9 days   2019-11-30   2019-12-09
2  xxxx/xxxxx.0183679721 2019-11-28 2019-11-28        NaT NaT 2 days 6 days   2019-11-30   2019-12-06
4  xxxx/xxxxx.0183888975 2019-11-20 2019-12-10        NaT NaT 6 days 1 days   2019-12-16   2019-12-17