CODEWITHSUNDEEP
pythonpandasrolling-computation
GFB999
GFB999

Reputation: 43

How to perform rolling sum from another category


I'm trying to perform some calculation to retrieve the rolling total of a category between two occurrences of another category.
I'm conscious it's not easy to describe by words.
So, here is the example of the input dataframe and the expected output

Input:

Date       Category  Value
2012-01-04        A     10
2012-01-06        A     20
2012-02-15        B    -10
2012-04-29        A      5
2012-04-30        A     70
2012-10-15        A     15
2012-10-16        B    -30
2012-11-19        B    -50

Expected output: Only the rows B but mentioning the rolling total of A since the previous occurrence of B

Date       Category  Value  Total_A_since_previous_B
2012-02-15        B    -10                        30
2012-10-16        B    -30                        90
2012-11-19        B    -50                         0

I've tried several things without success.
Can you help me to understand how to do that?

Upvotes: 4

Views: 149

Answers (2)

ansev
ansev

Reputation: 30920

Use GroupBy.agg:

blocks = df.Category.shift().eq('B').cumsum()
new_df = (df.groupby(blocks)
            .agg(Date= ('Date','last'),
                 Category = ('Category','last'),
                 Value = ('Value','last'),
                 Total_A_since_previous_B = ('Value','sum')
                )
             .assign(Total_A_since_previous_B = lambda x: x.Total_A_since_previous_B
                                                           .sub(x.Value))
             .reset_index(drop=True))

print(new_df)
         Date Category  Value  Total_A_since_previous_B
0  2012-02-15        B    -10                        30
1  2012-10-16        B    -30                        90
2  2012-11-19        B    -50                         0

Upvotes: 2

Allen Qin
Allen Qin

Reputation: 19947

First create groups for each occurence of B,sum the Values and then assign it as a new column to the filtered df.

(
    pd.Series(np.where(df.Category.eq('B'), df.index, np.nan)).bfill()
    .pipe(lambda x: df.groupby(x).Value.apply(lambda x: x[:-1].sum()))
    .pipe(lambda x: df[df.Category=='B'].assign(Total_A_since_previous_B=x))
)

    Date        Category    Value   Total_A_since_previous_B
2   2012-02-15  B           -10     30
6   2012-10-16  B           -30     90
7   2012-11-19  B           -50     0

Upvotes: 2

Related Questions