Odtnhj

Consider the following data that closely resembles the Pandas' Group By Tutorial:

import pandas as pd
import numpy as np

df = pd.DataFrame('Week' : [1, 2, 1, 2,
 1, 2, 1, 1],
 'BloodType' : ['A+', 'AB', 'AB', 'B',
 'B', 'B+', 'AB', 'AB'],
 'C' : np.random.randn(8),
 'D' : np.random.randn(8))

This produces a DataFrame that looks like this:

I want to group by the "Week" and then apply some operation to only the columns C and D. So I tried:

week_group = df.groupby('Week')
week_group.apply(lambda x: x.drop(["BloodType", "Week"], 1))

Which I originally interpreted as for every DataFrame drop the "BloodType" and "Week" column and give me the resulting group. However, it gives me:

However, I would have expected it to give me a Group, where each index was a DataFrame with only columns C and D. I did not expect a DataFrame.

I tried switching out apply with transform and agg which gave:

ValueError: transform must return a scalar value for each group

and:

ValueError: cannot copy sequence with size 2 to array axis with dimension 5

respectively. Is there a relatively simple transformation that can remove rows by name for each DataFrame in in a pandas Group and return the resulting Group object (or perform the operation in place)?

Based off of CJ59's answer I came up with this concise solution:

week_group = week_group[df.columns.difference(["Week", "BloodType"])]

Are you perhaps searching for

for name, group in df.groupby('Week'):
 print(name, group.drop(columns=['Week', 'BloodType']))

1 C D
0 0.496714 -0.469474
2 0.647689 -0.463418
4 -0.234153 0.241962
6 1.579213 -1.724918
7 0.767435 -0.562288
2 C D
1 -0.138264 0.54256
3 1.523030 -0.46573
5 -0.234137 -1.91328