15

I have a list as follows.

[['Andrew', '1', '9'], ['Peter', '1', '10'], ['Andrew', '1', '8'], ['Peter', '1', '11'], ['Sam', '4', '9'], ['Andrew', '2', '2']]

I would like sum up the last column grouped by the other columns.The result is like this

[['Andrew', '1', '17'], ['Peter', '1', '21'], ['Sam', '4', '9'], ['Andrew', '2', '2']]

which is still a list.

In real practice, I would always like to sum up the last column grouped by many other columns. Is there a way I can do this in Python? Much appreciated.

jpp
  • 159,742
  • 34
  • 281
  • 339
Deepleeqe
  • 317
  • 1
  • 8

6 Answers6

14

dynamically grouping by all columns except the last one:

In [24]: df = pd.DataFrame(data)

In [25]: df.groupby(df.columns[:-1].tolist(), as_index=False).agg(lambda x: x.astype(int).sum()).values.tolist()
Out[25]: [['Andrew', '1', 17], ['Andrew', '2', 2], ['Peter', '1', 21], ['Sam', '4', 9]]
MaxU - stand with Ukraine
  • 205,989
  • 36
  • 386
  • 419
10

This is an O(n) solution via collections.defaultdict, adaptable to any number of keys.

If your desired output is a list, then this may be preferable to a solution via Pandas, which requires conversion to and from a non-standard type.

from collections import defaultdict

lst = [['Andrew', '1', '9'], ['Peter', '1', '10'], ['Andrew', '1', '8'],
       ['Peter', '1', '11'], ['Sam', '4', '9'], ['Andrew', '2', '2']]

d = defaultdict(int)

for *keys, val in lst:
    d[tuple(keys)] += int(val)

res = [[*k, v] for k, v in sorted(d.items())]

Result

[['Andrew', '1', 17], ['Andrew', '2', 2], ['Peter', '1', 21], ['Sam', '4', 9]]

Explanation

  • Cycle through your list of lists, define keys / value and add to your defaultdict of lists.
  • Use a list comprehension to convert dictionary to desired output.
jpp
  • 159,742
  • 34
  • 281
  • 339
9

Op1

You can pass a index sum and add tolist convert back to list

pd.DataFrame(L).\
   set_index([0,1])[2].astype(int).sum(level=[0,1]).\
        reset_index().values.tolist()
Out[78]: [['Andrew', '1', 17], ['Peter', '1', 21], ['Sam', '4', 9], ['Andrew', '2', 2]]

Op2

For list of list you can using groupby from itertools

from itertools import groupby
[k+[sum(int(v) for _,_, v in g)] for k, g in groupby(sorted(l), key = lambda x: [x[0],x[1]])]
Out[98]: [['Andrew', '1', 17], ['Andrew', '2', 2], ['Peter', '1', 21], ['Sam', '4', 9]]
BENY
  • 317,841
  • 20
  • 164
  • 234
7

Create to DataFrame and aggregate third column converted to integers by first and second columns, last convert back to lists:

df = pd.DataFrame(L)
L = df[2].astype(int).groupby([df[0], df[1]]).sum().reset_index().values.tolist()
print (L)
[['Andrew', '1', 17], ['Andrew', '2', 2], ['Peter', '1', 21], ['Sam', '4', 9]]

And solution with defaultdict, python 3.x only:

from collections import defaultdict

d = defaultdict(int)
#https://stackoverflow.com/a/10532492
for *head, tail in L:
    d[tuple(head)] += int(tail)

d = [[*i, j] for i, j in sorted(d.items())]
print (d)
[['Andrew', '1', 17], ['Andrew', '2', 2], ['Peter', '1', 21], ['Sam', '4', 9]]
jezrael
  • 822,522
  • 95
  • 1,334
  • 1,252
6

pd.factorize and np.bincount

f, u = pd.factorize([tuple(x[:-1]) for x in data])
v = np.array([x[-1] for x in data], int)

[list(k) + [int(v)] for k, v in zip(u, np.bincount(f, v))]

[['Andrew', '1', 17], ['Peter', '1', 21], ['Sam', '4', 9], ['Andrew', '2', 2]]
piRSquared
  • 285,575
  • 57
  • 475
  • 624
0

Added my 2 cents. Both of these use groupby, agg

V1: introduces the new sum function.

sum=lambda x: x.astype(int).sum()
print(df.groupby([0,1], as_index=False).agg({2: sum}).values.tolist())

V2: converts the column to numeric

df[2] = pd.to_numeric(df[2])
print(df.groupby([0,1], as_index=False).agg({2: sum}).values.tolist())

And will return:

[['Andrew', '1', 17], ['Andrew', '2', 2], ['Peter', '1', 21], ['Sam', '4', 9]]
prosti
  • 42,291
  • 14
  • 186
  • 151