Python - 用列中的先前记录值填充NULL

Question

import pandas as pd

df = pd.DataFrame([['NewJersy',0,'2020-08-29'],
                   ['NewJersy',12,'2020-08-30'],
                   ['NewJersy',12,'2020-08-31'],
                   ['NewJersy',None,'2020-09-01'],
                   ['NewJersy',None,'2020-09-02'],
                   ['NewJersy',None,'2020-09-03'],
                   ['NewJersy',5,'2020-09-04'],
                   ['NewJersy',5,'2020-09-05'],
                   ['NewJersy',None,'2020-09-06'],
                   ['NewYork',None,'2020-08-29'],
                   ['NewYork',None,'2020-08-30'],
                   ['NewYork',8,'2020-08-31'],
                   ['NewYork',7,'2020-09-01'],
                   ['NewYork',None,'2020-09-02'],
                   ['NewYork',None,'2020-09-03']],
                   columns=['FName', 'FVal', 'GDate'])

print(df)

I want to fill NULL value with previous record values. For example Column FValue has values NULL for 20-09-01 to 20-09-03. The NULL value should be replaced with value 12 taken from previous valid value ie,from 20-08-31.

Also if the value for date 2020-08-29 is null then it should be replaced with zero as it is the first date and it doesn't have previous record.

I tried below code but not working

df['F'] = df['F'].fillna(method='ffill')

Check for Expected Values here: Fill Null Values image

Thanks

Answer 1

You should first ensure your DataFrame is sorted along time in case:

df = df.sort_values('GDate').reset_index(drop=True)

Then you must fill the first value with a 0:

if pd.isnull(df.loc[0, 'FVal']):
    df.loc[0, 'FVal'] = df.loc[0, 'FVal']

And then forward fill as you did:

df['FVal'] = df['FVal'].fillna(method='ffill')

Note that the column name is FVal not F .

Answer 2

Not sure if this is what you want. But this is what I would do

>>> import math
>>> for s in df.iterrows():
...     if math.isnan(s[1][1]):
...        df.iloc[s[0],1] = df.iloc[s[0] - 1,1]
...
>>> df
       FName  FVal       GDate
0   NewJersy   0.0  2020-08-29
1   NewJersy  12.0  2020-08-30
2   NewJersy  12.0  2020-08-31
3   NewJersy  12.0  2020-09-01
4   NewJersy  12.0  2020-09-02
5   NewJersy  12.0  2020-09-03
6   NewJersy   5.0  2020-09-04
7   NewJersy   5.0  2020-09-05
8   NewJersy   5.0  2020-09-06
9    NewYork   5.0  2020-08-29
10   NewYork   5.0  2020-08-30
11   NewYork   8.0  2020-08-31
12   NewYork   7.0  2020-09-01
13   NewYork   7.0  2020-09-02
14   NewYork   7.0  2020-09-03
>>>

Answer 3

You can try this:

df.GDate = pd.to_datetime(df.GDate)
for i in range(len(df)):
    if (np.isnan(df.FVal.loc[i])) and (i > 0):
        if (df.GDate.loc[i]-df.GDate.loc[i-1]).days == 1:
            print((df.GDate.loc[i]-df.GDate.loc[i-1]).days)
            df.FVal.loc[i] = df.FVal.loc[i-1]
        else:
            df.FVal.loc[i] = 0

Output

    FName       FVal    GDate
0   NewJersy    0.0     2020-08-29
1   NewJersy    12.0    2020-08-30
2   NewJersy    12.0    2020-08-31
3   NewJersy    12.0    2020-09-01
4   NewJersy    12.0    2020-09-02
5   NewJersy    12.0    2020-09-03
6   NewJersy    5.0     2020-09-04
7   NewJersy    5.0     2020-09-05
8   NewJersy    5.0     2020-09-06
9   NewYork     0.0     2020-08-29
10  NewYork     0.0     2020-08-30
11  NewYork     8.0     2020-08-31
12  NewYork     7.0     2020-09-01
13  NewYork     7.0     2020-09-02
14  NewYork     7.0     2020-09-03