Merge Pandas Dataframe Rows based on multiple conditions-CodePudding

Hi I have a pandas df which contains dates and amounts.

    Date    Amount  
0 10/02/22  1600       
1 10/02/22  150     
2 11/02/22  100       
3 11/02/22  800
4 11/02/22  125

If an entry is one day later and less than 10% of any other entry I would like to sum the amounts and then take the earliest date.

So the df would look like:

Date    Amount  
0 10/02/22  1825       
1 10/02/22  150         
2 11/02/22  800

I've tried creating threshold and then creating groups based on these conditions but this does not yield expected results.

threshold_selector =  (amount_difference < 0.1) & (date_difference == day)

Where day is a time delta of one day

groups = threshold_selector.cumsum()
dates= dates.groupby(groups).agg({'Amount':sum, 'Date': min})

The result is all rows joined into one.

CodePudding user response：

Here is an alternative using a groupby approach:

# ensure datetime
df['Date'] = pd.to_datetime(df['Date'], dayfirst=True)

# group Amounts by Date
g = df.groupby('Date')['Amount']

# get max amount per date
date_max = g.max()
# shift to previous date
prev_date_max = date_max.shift(1, freq='D').reindex(date_max.index, fill_value=0)

# identify rows to drop later
mask = df['Amount'].div(df['Date'].map(prev_date_max)).le(0.1)

# get value of next day to add to max
val_to_add = (df['Amount'][mask]
                 .groupby(df['Date']).sum()
                 .shift(-1, freq='D')
                )

# add to max
df['Amount']  = df['Date'].map(val_to_add).where(df.index.isin(g.idxmax())).fillna(0)

# drop rows
df = df.loc[~mask]

output:

        Date  Amount
0 2022-02-10  1825.0
1 2022-02-10   150.0
3 2022-02-11   800.0

CodePudding user response：

I would approach this using a pivot.

Sort the values with descending amount and pivot to have the largest value in the first column. Then find the values lower or equal to 10% that and mask them add to first column. Then shape back to original:

df['Date'] = pd.to_datetime(df['Date'], dayfirst=True)
df = df.sort_values(by=['Date', 'Amount'], ascending=[True, False])

# pivot to have col0 with the largest value per day
df2 = (df
 .assign(col=df.groupby('Date').cumcount())
 .pivot(index='Date', columns='col', values='Amount')
)

# identify values lower than the 10% of the previous day's max
mask = df2.div(df2[0].shift(1, freq='D'), axis=0).le(0.1).reindex_like(df2)

# add the lower than 10% values
df2[0]  = df2.where(mask).sum(axis=1).shift(-1, 'D').reindex(mask.index, fill_value=0)

# mask them
df2 = df2.mask(mask)

# reshape back dropping the NaNs
df2 = df2.stack().droplevel('col').reset_index(name='Amount')

output:

        Date  Amount
0 2022-02-10  1825.0
1 2022-02-10   150.0
2 2022-02-11   800.0