Numpy: Adding specific columns of rows conditionally

Question

For a given numpy array:

[[1, 1, 'IGNORE_THIS_COL', 100],
 [1, 1, 'IGNORE_THIS_COL', 101],
 [1, 2, 'IGNORE_THIS_COL', 100]]

Is it possible to sum the rows (and columns conditionally)? Say column 0 is group and column 1 is user, then I would like to add the fourth column accordingly. The final 'summed' array should look like this.

[[1, 1, 'IGNORE_THIS_COL', 201],
 [1, 2, 'IGNORE_THIS_COL', 100]]

I have already checked multiple answers, including Numpy: conditional sum.

@user3483203 In this case it is an integer. Does that change the solution? — DaveIdito, Jun 30 '18 at 19:45
Very much so, otherwise numpy would cast all to strings when you created the array — user3483203, Jun 30 '18 at 19:45

score 1 · Accepted Answer · answered Jun 30 '18 at 19:39

1

You're looking for a groupby on a subset of columns. This is a challenge to implement with numpy, but is straightforward with a pandas groupby:

import pandas as pd

df = pd.DataFrame(array)
out = df.groupby([0, 1], as_index=False).agg({2:'first', 3:'sum'}).values.tolist()

print(out)
[[1, 1, 'IGNORE_THIS_COL', 201], [1, 2, 'IGNORE_THIS_COL', 100]]

answered Jun 30 '18 at 19:39

cs95

379,657
97
704
746

What is the purpose of `2:'first'` in the aggregation? – DaveIdito Jun 30 '18 at 20:07
1

@DaveIdito You wanted to ignore the column, so I'm ignoring it by just taking the first value from it per group. – cs95 Jun 30 '18 at 20:08

Numpy: Adding specific columns of rows conditionally

1 Answers1