为每个日期维护默认值 True/False

Question

. Answers to this question are eligible for a +200 reputation bounty. Digital Farmer wants to to this question.

During the day, new investment possibilities are registered, but the results ( lay column) are only registered at midnight each day.

So let's assume this CSV :

clock_now,competition,market_name,lay
2022/12/30,A,B,-1
2022/12/31,A,B,1.28
2023/01/01,A,B,-1
2023/01/02,A,B,1
2023/01/03,A,B,1
2023/01/04,A,B,
2023/01/04,A,B,
2023/01/04,A,B,

Until yesterday, 2023/01/03 , the sum of the lines that have the value A in competition and B in market_name , was +1.28

I only invest if it is above 0 , so during today, every time this combination of values comes, the answer will be True to invest.

At the end of the day, when the lay values are registered, I look at the total result:

clock_now,competition,market_name,lay
2022/12/30,A,B,-1
2022/12/31,A,B,1.28
2023/01/01,A,B,-1
2023/01/02,A,B,1
2023/01/03,A,B,1
2023/01/04,A,B,-1
2023/01/04,A,B,-1
2023/01/04,A,B,-1

End of the day: -1,72

This means that tomorrow, if that same combination of values appears in the columns, I will not invest once because it will always be negative because it only calculates the values that it has until the previous day.

I'm trying to create a column to show where it was True and where it was False:

df = pd.read_csv('example.csv')
combinations = [['market_name', 'competition']]
for cbnt in combinations:
    df['invest'] = (df.groupby(cbnt)['lay']
                      .apply(lambda s: s.cumsum().shift())
                      .gt(df['lay'])
                   )

    df['cumulative'] = (df.groupby(cbnt)['lay']
                      .apply(lambda s: s.cumsum().shift())
                   )

    print(df[['clock_now','invest','cumulative']])

But the result is this:

    clock_now  invest  cumulative
0  2022/12/30   False         NaN
1  2022/12/31   False       -1.00
2  2023/01/01    True        0.28
3  2023/01/02   False       -0.72
4  2023/01/03   False        0.28
5  2023/01/04    True        1.28
6  2023/01/04    True        0.28
7  2023/01/04    True       -0.72

The expected result would be this:

    clock_now  invest  cumulative
0  2022/12/30   False         NaN
1  2022/12/31   False       -1.00
2  2023/01/01    True        0.28
3  2023/01/02   False       -0.72
4  2023/01/03    True        0.28
5  2023/01/04    True        1.28
6  2023/01/04    True        0.28
7  2023/01/04    True       -1.72

How should I proceed so that cumsum can understand that attention must be paid to maintaining a daily pattern according to the results of previous days?

Example Two:

clock_now,competition,market_name,lay
2022/08/09,A,B,-1.0
2022/08/12,A,B,1.28
2022/09/07,A,B,-1.0
2022/10/15,A,B,1.0
2022/10/15,A,B,-1.0
2022/11/20,A,B,1.0

Note that on 2022/10/15 , it is delivering one False and one True , so in fact it is not tracking according to the date which is how I want it to happen:

    clock_now  invest  cumulative
0  2022/08/09   False         NaN
1  2022/08/12   False       -1.00
2  2022/09/07    True        0.28
3  2022/10/15   False       -0.72
4  2022/10/15    True        0.28
5  2022/11/20   False       -0.72

The correct would be always or all False or all True when on equal dates. Like this:

    clock_now  invest  cumulative
0  2022/08/09   False         NaN
1  2022/08/12   False       -1.00
2  2022/09/07    True        0.28
3  2022/10/15   False       -0.72
4  2022/10/15   False        0.28
5  2022/11/20   False       -0.72

Answer 1

(df.join(
    # Count market&competition specific cumsum for each row
    # and join back with df
    df.groupby(['market_name', 'competition']).lay.cumsum().rename('lay_cumsum') > 0
)
# Group by market&comp&date to get last cumsum within each day
.groupby(['market_name', 'competition', 'clock_now']) 
# Get cumsum Series for each group
.lay_cumsum
# Getting last cumsum within group
.last()
# Group by market&comp
.groupby(['market_name', 'competition'])
# Shift by one to assign to each date prev date's cumsum
.shift(1)
.rename('lay_cumsum')
.reset_index()
# Merge back with original df
.merge(df, on=['clock_now', 'market_name', 'competition']))

This will output

  market_name competition   clock_now lay_cumsum   lay
0           B           A  2022/12/30        NaN -1.00
1           B           A  2022/12/31      False  1.28
2           B           A  2023/01/01       True -1.00
3           B           A  2023/01/02      False  1.00
4           B           A  2023/01/03       True  1.00
5           B           A  2023/01/04       True -1.00
6           B           A  2023/01/04       True -1.00
7           B           A  2023/01/04       True -1.00