如何在熊猫数据框中将列表作为输入作为groupby函数的输入

Question

Suppose a subset of a dataset comprises these 2 columns,

     attacker_king              attacker_commander
0   Joffrey/Tommen Baratheon    Jaime Lannister
1   Joffrey/Tommen Baratheon    Gregor Clegane
2   Joffrey/Tommen Baratheon    Jaime Lannister, Andros Brax
3   Robb Stark                  Roose Bolton, Wylis Manderly, Medger Cerwyn
4   Robb Stark                  Robb Stark, Brynden Tully
5   Robb Stark                  Robb Stark, Tytos Blackwood, Brynden Tully

My objective is to get the 'set of commanders' that each king deploys, as per the dataset.

[x for x in battles['attacker_commander'].dropna().str.split(',').sum()]

The above command obtains only comma separated list of commanders But if I choose to use the following list comprehension,

battles[['attacker_commander','attacker_king']].groupby('attacker_king').sum()

I get an output where

attacker_king                      attacker_commander   
Balon/Euron Greyjoy         Victarion GreyjoyAsha GreyjoyTheon GreyjoyTheo...
Joffrey/Tommen Baratheon    Jaime LannisterGregor CleganeJaime Lannister, ...
Robb Stark                  Roose Bolton, Wylis Manderly, Medger Cerwyn, H...
Stannis Baratheon           Stannis Baratheon, Davos SeaworthStannis Barat...

The problem with this approach is, suppose a row has just 1 commander ,when that is summed with next row, output can look like 'Victarion GreyjoyAsha Greyjoy' instead of 'Victarion Greyjoy,Asha Greyjoy'. So does it make sense to use the list created using

[x for x in battles['attacker_commander'].dropna().str.split(',').sum()]

and feed it to a groupby('attacker_king') or what approach do you folks suggest?

Answer 1

I think you need apply with function join first:

battles.groupby('attacker_king')['attacker_commander'].apply(','.join)

If need remove NaN :

battles.groupby('attacker_king')['attacker_commander'].apply(lambda x: ','.join(x.dropna()))

Then split and use set for unique values:

df = battles.groupby('attacker_king')['attacker_commander']
            .apply(lambda x: list(set(','.join(x.dropna()).split(','))))
print (df)

The best solution for debugging is use custom function and then rewrite code to lambda :

def f(x):
    #Series by attacker_commander per group
    print (x)
    #first remove NaN
    print (x.dropna())
    #join by ,
    print (','.join(x.dropna()))
    #create list by split
    print (','.join(x.dropna()).split(','))
    #convert to set - unique values
    print (set(','.join(x.dropna()).split(',')))
    #set convert to list
    print (list(set(','.join(x.dropna()).split(','))))
    return list(set(','.join(x.dropna()).split(',')))

df = battles.groupby('attacker_king')['attacker_commander'].apply(f)
print (df)

---

But also one posssible solution is remove rows with NaN by column DataFrame.dropna first:

def f(x):
    return list(set(','.join(x).split(',')))

df = battles.dropna(subset=['attacker_commander']).groupby('attacker_king')['attacker_commander'].apply(f)
print (df)

Answer 2

you want to join the strings by groups then split and find the unique values.

df.groupby(
    'attacker_king'
).attacker_commander.apply(','.join).str.split(',').apply(pd.unique)

attacker_king
Joffrey/Tommen Baratheon      [Jaime Lannister, Gregor Clegane,  Andros Brax]
Robb Stark                  [Roose Bolton,  Wylis Manderly,  Medger Cerwyn...
Name: attacker_commander, dtype: object