This question is a variation of this one, the only difference being that the dates can repeat in the rows of a DataFrame. So, the example would be:
| Date | Position | TrainerID | Win% |
|---|---|---|---|
| 2017-09-03 | 4 | 1788 | 0 (0 wins, 1 race) |
| 2017-09-16 | 5 | 1788 | 0 (0 wins, 2 races) |
| 2017-10-14 | 1 | 1788 | 33 (1 win, 3 races) |
| 2017-10-14 | 3 | 1788 | 25 (1 win, 4 races) |
Is it possible to compute Win% in these conditions for the last 1000 days? If so, how?
The logic in the other solution is still correct; the issue is that groupby + rolling destroys the Index so it becomes problematic to align the result back to the original DataFrame.
In this case, you can .reset_index and use max (assuming a RangeIndex) to bring along the original Index. This allows you to aggregate, then align the result back.
I added an additional row at the end to show you how it enforces the 1000 day window.
# If your DataFrame doesn't have a RangeIndex this is required for the logic
#df = df.reset_index(drop=True)
df['win'] = df['Position'].eq(1)
s = (df.reset_index().groupby('TrainerID')
.rolling('1000D', on='Date')
.agg({'win': 'mean', 'index': 'max'})
.reset_index(drop=True)
.set_index('index')
.mul(100))
# win
#index
#0.0 0.000000
#1.0 0.000000
#2.0 33.333333
#3.0 25.000000
#4.0 100.000000
df['Win %'] = s
print(df)
Date Position TrainerID win Win %
0 2017-09-03 4 1788 False 0.000000
1 2017-09-16 5 1788 False 0.000000
2 2017-10-14 1 1788 True 33.333333
3 2017-10-14 3 1788 False 25.000000
4 2027-10-14 1 1788 True 100.000000