对两个熊猫数据框列求和，保留不常见的行

Question

I just asked a similar question but then realized, it wasn't the right question.

What I'm trying to accomplish is to combine two data frames that actually have the same columns, but may or may not have common rows (indices of a MultiIndex). I'd like to combine them taking the sum of one of the columns, but leaving the other columns.

According to the accepted answer, the approach may be something like:

def mklbl(prefix,n):
    try:
        return ["%s%s" % (prefix,i)  for i in range(n)]
    except:
        return ["%s%s" % (prefix,i) for i in n]



mi1 = pd.MultiIndex.from_product([mklbl('A',4), mklbl('C',2)])

mi2 = pd.MultiIndex.from_product([mklbl('A',[2,3,4]), mklbl('C',2)])

df2 = pd.DataFrame({'a':np.arange(len(mi1)), 'b':np.arange(len(mi1)),'c':np.arange(len(mi1)), 'd':np.arange(len(    mi1))[::-1]}, index=mi1).sort_index().sort_index(axis=1)    

df1 = pd.DataFrame({'a':np.arange(len(mi2)), 'b':np.arange(len(mi2)),'c':np.arange(len(mi2)), 'd':np.arange(len(    mi2))[::-1]}, index=mi2).sort_index().sort_index(axis=1)


df1 = df1.add(df2.pop('b'))

but the problem is this will fail as the indices don't align.

This is close to what I'm trying to achieve, except that I lose rows that are not common to the two dataframes:

df1['b'] = df1['b'].add(df2['b'], fill_value=0)

But this gives me:

Out[197]: 
       a   b  c  d
A2 C0  0   4  0  5
   C1  1   6  1  4
A3 C0  2   8  2  3
   C1  3  10  3  2
A4 C0  4   4  4  1
   C1  5   5  5  0

When I want:

In [197]: df1
Out[197]: 
       a   b  c  d
A0 C0  0  0  0  7
   C1  1  2  1  6
A1 C0  2  4  2  5
   C1  3  6  3  4
A2 C0  0   4  0  5
   C1  1   6  1  4
A3 C0  2   8  2  3
   C1  3  10  3  2
A4 C0  4   4  4  1
   C1  5   5  5  0

Note: in response to @RandyC's comment about the XY problem... the specific problem is that I have a class which reads data and returns a dataframe of 1e9 rows. The columns of the data frame are latll, latur, lonll, lonur, concentration, elevation . The data frame is indexed by a MultiIndex (lat, lon, time) where time is a datetime. The rows of the two dataframes may/may not be the same (IF they exist for a given date, the lat/lon will be the same... they are grid cell centers). latll, latur, lonll, lonur are calculated from lat/lon. I want to sum the concentration column as I add two data frames, but not change the others.

Answer 1

Self answering, there was an error in the comment above that caused a double adding. This is correct:

newdata = df2.pop('b')
result = df1.combine_first(df2)
result['b']= result['b'].add(newdata, fill_value=0)

seems to provide the solution to my use-case.