在pandas Dataframe中查找倒数行

Question

I have this dataframe, and need to retain only those lines having a reciprocal values for 2 columns (numA and numB here).

gpm = pd.DataFrame(data={
    'id':[1,2,3,4,5,6,7,8,9],
    'time':[150315,150315,150315,150315,150315,150315,150315,150315,150315],
    'numA':['A','D','C','B','A','C','A','E','D'],
    'numB':['B','C','B','A','B','D','B','A','A'],
    'antA':['MSPDV','VIELU','RMPC1','MJCIH','PALT2','M2PV3','MACIF','MACIF','VIELU'],
    'antB':['BPDV8','0GRI3','SSFDJ','SSFDJ','SSFDJ','CCPG1','0GRI3','SSFDJ','SSFDJ']
    })

I only want lines in which columns numA and numB are reciprocal. That is, retaining al lines where the pairs (A,B), (B,A) and (C,D),(D,C) occur.

My solution, for now, involves making a list of all unique identifiers and going through each line looking whether the actual partner is in the list of partners

it is extremely slow.... (and perhaps incorrect!)

## here's my code
parties = {}
nums = gpm['numA']+gpm['numB']
for i in nums.unique():
    parties[i] = gpm['numB'][gpm['numA'] == i]
    parties[i] = gpm['numA'][gpm['numB'] == i]

new_d = gpm.iloc[[0]]
for i in np.arange(1,gpm.shape[0]):
    numa = gpm.iloc[i]['numA']
    if gpm.iloc[i]['numB'] in parties[numa]:
        new_d.append(gpm.iloc[[i]])

any savvy coder that could help speed this up? The actual file to parse is a ~15GB csv.

Thanks

Answer 1

In your example, I assume the rows with id=3, 8 & 9, which are (C, B), (E, A) and (D, A), are unwanted ? If so, here's a standard way to select by comparing the values in numA and numB for specific acceptable combinations:

In [5]: gpm[((gpm['numA'] == 'A') & (gpm['numB'] == 'B')) |
   ...:     ((gpm['numA'] == 'B') & (gpm['numB'] == 'A')) |
   ...:     ((gpm['numA'] == 'C') & (gpm['numB'] == 'D')) | 
   ...:     ((gpm['numA'] == 'D') & (gpm['numB'] == 'C'))
   ...: ]
Out[5]:
   id    time numA numB   antA   antB
0   1  150315    A    B  MSPDV  BPDV8
1   2  150315    D    C  VIELU  0GRI3
3   4  150315    B    A  MJCIH  SSFDJ
4   5  150315    A    B  PALT2  SSFDJ
5   6  150315    C    D  M2PV3  CCPG1
6   7  150315    A    B  MACIF  0GRI3

(assign the result of that to new_d )