apply lambda function using group by and using previous row value

Question

I have the following data frame, per each date, per hour, I want create a new column "result"such that if the value in column "B" is >=0 then use the value in column A; otherwise use the maximum between 0 and the previous row value in column B

Date    Hour    A     B    result
1/1/2018    1    5     95    5
1/1/2018    1    16    79    16
1/1/2018    1    85   -6     79
1/1/2018    1    12   -18    0
1/1/2018    2    17    43    17
1/1/2018    2    17    26    17
1/1/2018    2    16    10    16
1/1/2018    2    142  -132   10
1/1/2018    2    10   -142   0

I tried grouping by date and hour and then applying a lambda function using shift but I got an error:

df['result'] = df.groupby(['Date','Hour']).apply(lambda x: x['A'] if x['B'] >= 0 else np.maximum(0, x['B'].shift(1)), axis = 1)

Answer 1

Use np.where. The groupby is only necessary when shifting "B", so you can vectorise this operation without using apply.

df['result'] = np.where(
    df.B >= 0, 
    df.A, 
    df.groupby(['Date', 'Hour'])['B'].shift().clip(lower=0))
df

       Date  Hour    A    B  result
0  1/1/2018     1    5   95     5.0
1  1/1/2018     1   16   79    16.0
2  1/1/2018     1   85   -6    79.0
3  1/1/2018     1   12  -18     0.0
4  1/1/2018     2   17   43    17.0
5  1/1/2018     2   17   26    17.0
6  1/1/2018     2   16   10    16.0
7  1/1/2018     2  142 -132    10.0
8  1/1/2018     2   10 -142     0.0