填补 pandas df 中的空白

Question

I have the below df:

import pandas as pd
df = pd.DataFrame(
{"col1": [2000, 2000, 2000, '', 2001, 2001, '', '', 2002],
 "col2": ["b1", "c1", "d1", '' , "c1", "d1", '', '', "d1"],
 "col3": [10, 20, 30, '', 20, 40, '', '', 60]  
}
)
df

    col1    col2    col3
0   2000    b1       10
1   2000    c1       20
2   2000    d1       30
3           
4   2001    c1       20
5   2001    d1       40
6           
7           
8   2002    d1       60

I need 3 rows for each date from 2000 to 2002 and each date will have b1, c1 and d1. When a row is missing (like rows 3, 6 and 7) I want to fill it so that it has a date, a b1, c1 or d1 and col3 will be 0 just like in df2 below:

df2 = pd.DataFrame(
{"col1": [2000, 2000, 2000, 2001, 2001, 2001, 2002, 2002, 2002],
 "col2": ["b1", "c1", "d1", "b1" , "c1", "d1", "b1", "c1", "d1"],
 "col3": [10, 20, 30, 0, 20, 40, 0, 0, 60]  
}
)
df2

    col1    col2    col3
0   2000    b1      10
1   2000    c1      20
2   2000    d1      30
3   2001    b1      0
4   2001    c1      20
5   2001    d1      40
6   2002    b1      0
7   2002    c1      0
8   2002    d1      60

How to do this in pandas? (I have a large dataframe like this with many dates not just 3 but this example will help me get the idea!)

Answer 1

You can take a cartesian product of the expected values with year and create possibilities.

Then merge(left join) and fillna:

df = df.replace('',np.nan).dropna(subset=['col1'])
rows = ['b1','c1','d1']
possibilities = pd.MultiIndex.from_product((df['col1'].unique(),rows))

out = (pd.DataFrame(possibilities.tolist(),columns=['col1','col2'])
       .merge(df,how='left').fillna({"col3":0},downcast='infer'))
out['col1']= out['col1'].astype(int)

Or:

out = (possibilities.to_frame(name=['col1','col2']).merge(df,how='left')
      .fillna({"col3":0},downcast='infer'))
out['col1']= out['col1'].astype(int)

---

print(out)

   col1 col2  col3
0  2000   b1    10
1  2000   c1    20
2  2000   d1    30
3  2001   b1     0
4  2001   c1    20
5  2001   d1    40
6  2002   b1     0
7  2002   c1     0
8  2002   d1    60

Answer 2

Use DataFrame.reindex for add 0 for not existed combinations:

df = df2.replace('',np.nan).dropna(subset=['col1'])
rows = ['b1','c1','d1']
mux = pd.MultiIndex.from_product((df['col1'].unique(),rows), names=['col1','col2'])

df = df2.set_index(['col1','col2']).reindex(mux, fill_value=0).reset_index()
print (df)
   col1 col2  col3
0  2000   b1    10
1  2000   c1    20
2  2000   d1    30
3  2001   b1     0
4  2001   c1    20
5  2001   d1    40
6  2002   b1     0
7  2002   c1     0
8  2002   d1    60

Answer 3

First fill the original empty string with NaN

df = df.replace('', np.nan)

Then create a dummy dataframe from

dummy = pd.DataFrame([[x, y] for x in df['col1'].dropna().unique() for y in df['col2'].dropna().unique()], columns=['col1', 'col2'])

# You can also try multi index
# mux = pd.MultiIndex.from_product((df['col1'].dropna().unique(), df['col2'].dropna().unique()), names=['col1','col2'])
# dummy = pd.DataFrame({'col3': [0]*len(mux)}, index=mux).reset_index().reset_index()

print(dummy)

     col1 col2
0  2000.0   b1
1  2000.0   c1
2  2000.0   d1
3  2001.0   b1
4  2001.0   c1
5  2001.0   d1
6  2002.0   b1
7  2002.0   c1
8  2002.0   d1

At last, update NaN values in your original dataframe with dummy dataframe.

df.update(dummy, overwrite=False)
df.fillna(0, inplace=True)

print(df)

     col1 col2  col3
0  2000.0   b1  10.0
1  2000.0   c1  20.0
2  2000.0   d1  30.0
3  2001.0   b1   0.0
4  2001.0   c1  20.0
5  2001.0   d1  40.0
6  2002.0   b1   0.0
7  2002.0   c1   0.0
8  2002.0   d1  60.0