Pandas：如何将 MultiIndex DataFrame 与单个索引 DataFrame 连接起来，以及自定义排序

Question

I have a MultiIndex pandas DataFrame df_multi like:

import pandas as pd

df_multi = pd.DataFrame([['A', 'A1', 0,234,2002],['A', 'A1', 1,324,2550],
['A', 'A1', 2,345,3207],['A', 'A1', 3,458,4560],['A', 'A2', 0,569,1980],
['A', 'A2', 1,657,2314],['A', 'A2', 2,768,4568],['A', 'A2', 3,823,5761]], 
columns=['Product','Scenario','Time','Quantity','Price']).set_index(
['Product', 'Scenario'])

and a single index DataFrame df_single like:

df_single = pd.DataFrame([['A', -3,100],['A', -2,100], ['A', -1,100]],
columns=['Product','Time','Quantity']).set_index(['Product'])

For every 'Product' in the first index level of df_multi , and for every 'Scenario' in its second level, I would like to append/concatenate the rows in df_single , which contain some negative 'Time' values to be appended before the positive 'Time' values in df_multi begin.

I would furthermore like the resulting DataFrame to be first MultiIndexed by ['Product','Scenario'] (just like df_multi ), then secondly with the rows ordered by ascending value of 'Time'. In other words, the desired result is:

df_result = pd.DataFrame([['A', 'A1', -3,100,'NaN'],['A', 'A1', -2,100,'NaN'],
['A', 'A1', -1,100,'NaN'],['A', 'A1', 0,234,2002],['A', 'A1', 1,324,2550],
['A', 'A1', 2,345,3207],['A', 'A1', 3,458,4560],['A','A2', -3,100,'NaN'],
['A', 'A2', -2,100,'NaN'],['A', 'A2', -1,100,'NaN'],['A', 'A2', 0,569,1980],
['A', 'A2', 1,657,2314],['A', 'A2', 2,768,4568],['A', 'A2', 3,823,5761]],
columns=['Product','Scenario','Time','Quantity','Price']).set_index(
['Product', 'Scenario'])

EDIT:

• df_single has no 'Scenario' values, which can be confusing. As long as 'Product' matches, the same rows of df_single are to be appended to every scenario in df_multi , and they simply "inherit" the Scenario values for free.
• The actual DataFrames I'm working with are rather large (few thousand 'Product', few thousand 'Scenario' per product, and a few hundred 'Time' steps per scenario, plus extra columns which I did not write in the example), so I need to do this in a fully automated (and hopefully fast) way.

I tried to implement this with all of join , concat and merge , and I did not succeed. What would be the best way of achieving the desired result?

Answer 1

Consider resetting indexes as columns for a merge , followed by a groupby aggregation only to return one occurrence per group and avoid duplicates. Afterwards, run a concatenation, concat , followed by column sorting and setting back the multi-index.

# MERGE AND AGGREGATION
df_temp = df_multi.reset_index().merge(df_single.reset_index(), on='Product', suffixes=['','_'])\
                                .groupby(['Product', 'Scenario', 'Time_'])['Quantity_'].max()\
                                .reset_index().rename(columns={'Time_':'Time','Quantity_':'Quantity'})

# ROW BIND CONCATENATION
df_final = pd.concat([df_multi.reset_index(), df_temp])\
                    .sort_values(['Product','Scenario', 'Time'])\
                    .set_index(['Product', 'Scenario'])[['Time', 'Quantity', 'Price']]
print(df_final)
#                   Time  Quantity   Price
# Product Scenario                        
# A       A1          -3       100     NaN
#         A1          -2       100     NaN
#         A1          -1       100     NaN
#         A1           0       234  2002.0
#         A1           1       324  2550.0
#         A1           2       345  3207.0
#         A1           3       458  4560.0
#         A2          -3       100     NaN
#         A2          -2       100     NaN
#         A2          -1       100     NaN
#         A2           0       569  1980.0
#         A2           1       657  2314.0
#         A2           2       768  4568.0
#         A2           3       823  5761.0