Python Pandas：使用滚动回溯窗口计算累计总和

Question

I am calculating the value for the Total Wins (2 Days) column in the (table below) - also see below for comma separated values.

Total Wins (2 Days) is a cumulative count of the number of races the athlete has won either on that given day (eg. Day 5) or the day before (eg. Day 4) - as such, it is a count of the number of wins within a 2 day look back window. (I may want to change the look back window to any number).

For example, on Day 7: Jane gets a count of 1 because she won on Day 7 but lost on Day 6; Bill gets a count of 1 because he lost on Day 7 but won on Day 6. Steve didn't win on either day.

On Day 6, Bill gets a count of 1 because he won on day Day but lost on Day 5. Steve gets a count of 1 because he lost on Day 6 but won on Day 5. Jane won on neither day.

What is the best way to calculate 'Total Wins (2 Days)' in Python?

Follow-up Question

Also, as a follow-up question: With regards to the param for '.rolling(2)' (ie. 2 in this case), how would I set the param to be a value that is derived from other columns in the table?

What I want to do is change Race Day to a datetime object (see updated table below) and calculate 'Total Wins (X Days)' as the number of races won over the last week, month, year and so on. (The database I am working with is bigger than the example above).

I would prefer to define Year as a calendar year (ie. races won between 2014-01-01 and 2015-01-01), rather than simply 265 days.

Race Day,Athlete,Position,Total Wins,Total Wins (2 Days)
1,Steve,1,1,1
1,Jane,2,0,0
1,Bill,3,0,0
2,Bill,1,1,1
2,Steve,2,1,1
2,Jane,3,0,0
3,Jane,1,1,1
3,Bill,2,1,1
3,Steve,3,1,0
4,Steve,1,2,1
4,Jane,2,1,1
4,Bill,3,1,0
5,Steve,1,3,2
5,Jane,2,1,0
5,Bill,3,1,0
6,Bill,1,2,1
6,Steve,2,3,1
6,Jane,3,1,0
7,Jane,1,2,1
7,Bill,2,2,1
7,Steve,3,3,0

Race Day,Athlete,Position,Total Wins,Total Wins (2 Months)
2000-01-01,Steve,1,1,?
2000-01-01,Jane,2,0,?
2000-01-01,Bill,3,0,?
2000-01-02,Bill,1,1,?
2000-01-02,Steve,2,1,?
2000-01-02,Jane,3,0,?
2000-01-03,Jane,1,1,?
2000-01-03,Bill,2,1,?
2000-01-03,Steve,3,1,?
2000-01-04,Steve,1,2,?
2000-01-04,Jane,2,1,?
2000-01-04,Bill,3,1,?
2000-01-05,Steve,1,3,?
2000-01-05,Jane,2,1,?
2000-01-05,Bill,3,1,?
2000-01-06,Bill,1,2,?
2000-01-06,Steve,2,3,?
2000-01-06,Jane,3,1,?
2000-01-07,Jane,1,2,?
2000-01-07,Bill,2,2,?
2000-01-07,Steve,3,3,?

Answer 1

Create a Won column that captures Position 1 for each row and then apply rolling sum

df['Won'] = (df['Position'] == 1).astype(int)

df['Total Wins (2 Days)'] = df.groupby('Athlete')['Won'].apply(lambda x: x.rolling(2).sum()).combine_first(df['Won'])

    Race Day    Athlete Position    Total Wins  Total Wins (2 Days) Won
0   1           Steve   1           1           1.0                 1
1   1           Jane    2           0           0.0                 0
2   1           Bill    3           0           0.0                 0
3   2           Bill    1           1           1.0                 1
4   2           Steve   2           1           1.0                 0
5   2           Jane    3           0           0.0                 0
6   3           Jane    1           1           1.0                 1
7   3           Bill    2           1           1.0                 0
8   3           Steve   3           1           0.0                 0
9   4           Steve   1           2           1.0                 1
10  4           Jane    2           1           1.0                 0
11  4           Bill    3           1           0.0                 0
12  5           Steve   1           3           2.0                 1
13  5           Jane    2           1           0.0                 0
14  5           Bill    3           1           0.0                 0
15  6           Bill    1           2           1.0                 1
16  6           Steve   2           3           1.0                 0
17  6           Jane    3           1           0.0                 0
18  7           Jane    1           2           1.0                 1
19  7           Bill    2           2           1.0                 0
20  7           Steve   3           3           0.0                 0

You can delete the column Won using

df.drop('Won', axis = 1, inplace = True)

Option 2:

df['Won'] = (df['Position'] == 1).astype(int)

df['Total Wins (2 Days)'] = df.groupby('Athlete')['Won'].apply(lambda x: x.add(x.shift())).combine_first(df['Won'])

Edit: For the follow up question with Race Day being date, you can add a column by aggregating the data on day, week, month etc (based on your requirement) and then find the sum of current and previous period.

df['Race Day'] = pd.to_datetime(df['Race Day'])

df['Won'] = (df['Position'] == 1).astype(int)


df['Total Wins Day']=df.groupby(['Athlete', df['Race Day'].dt.day])['Won'].transform('sum')
df['Total Wins week']=df.groupby(['Athlete', df['Race Day'].dt.week])['Won'].transform('sum')
df['Total Wins Month']=df.groupby(['Athlete', df['Race Day'].dt.month])['Won'].transform('sum')

Now if you want total wins in last two months, you would use the Total Wins Month column and sum that with the previous column

df['Total Wins (2 Months)'] = df.groupby('Athlete')['Total Wins Month'].apply(lambda x: x.add(x.shift())).combine_first(df['Won'])