使用sort_values按y排序，使用drop_duplicates仅保留每个cust_id的一次出现：

out = df.sort_values('y', ascending=False).drop_duplicates('cust_id')
print(out)

# Output
   group_id  cust_id  score x1  x2  contract_id   y
0       101        1     95  F  30            1  30
3       101        2     85  M  28            2  18

正如@ifly6所建议的，你可以使用groupby_idxmax：

out = df.loc[df.groupby('cust_id')['y'].idxmax()]
print(out)

# Output
   group_id  cust_id  score x1  x2  contract_id   y
0       101        1     95  F  30            1  30
3       101        2     85  M  28            2  18

你可以使用drop_duplicates。下面是一个例子：

import pandas as pd
# Some data
df = pd.DataFrame({'cust_id': [1, 2, 1, 3, 4], 'y': [3, 4, 1, 5, 7]})
# Sorting by cust_id is actually not necessary
df.sort_values(by=['cust_id', 'y'], ascending=[True, True], inplace=True)
# Remove all the duplicates by cust_id, keeping the first one found
df.drop_duplicates(subset='cust_id', keep='last', inplace=True)

print(df)

    cust_id y
0   1   3
1   2   4
3   3   5
4   4   7