使用 Python pandas 执行索引匹配

Question

I'm struggling write some code to obtain the following functionality:

df1

Date          A      B
01/01/2021    39     100
01/02/2021    58     188
01/03/2021    220    300
01/04/2021    0      11

df2

Date          A      A      A      B     B     B
              0      50     100    0     100   200
01/01/2021    0.1    0.2    0.3    0.3   0.3   0.6
01/02/2021    0.1    0.2    0.3    0.3   0.3   0.6
01/03/2021    0.3    0.3    0.6    0.5   0.4   0.8
01/04/2021    0.3    0.3    0.6    0.5   0.8   0.8

df3 (desired output)

Date          A           B
01/01/2021    (39*0.1)    (100*0.3)
01/02/2021    (58*0.2)    (188*0.3)
01/03/2021    (220*0.6)   (300*0.8)
01/04/2021    (0*0.1)     (11*0.5)

Effectively, I need to check the values for A and B in df1 and multiply with the corresponding value in df2 based on date and whether the value is between 0 and 50, 50 and 100 or >100 (in the case of A).

In reality, df1 and df2 extend far beyond 2 items 'A' and 'B' and I intend to iterate for each column of df1 in a for loop thus I am looking for a general solution.

Thanks

Answer 1

Here is a way:

def fun(x):
   col_name = x.name
   col_idx = df.columns.get_loc(col_name)
   
   lower, middle, upper = df2.columns.get_loc_level(col_name)[1]
   cond_list = [upper <= x, middle <= x, lower <= x]
   choice_list =  np.arange(3)[::-1] + 3 * col_idx
   selections = np.select(cond_list, choice_list)

   cols_in_df2 = df2.columns[selections]
   rows_in_df2 = x.index
   multipliers_in_df2 = np.diag(df2.loc[rows_in_df2, cols_in_df2])
   
   result = x * multipliers_in_df2
   return result

df.apply(fun)

to get

                A      B
Date
2021-01-01    3.9   30.0
2021-01-02   11.6   56.4
2021-01-03  132.0  240.0
2021-01-04    0.0    5.5

We have a cond_list of conditions whose limits are obtained from the corresponding values in df2 s column, and a choice_list which is the corresponding columns in df2 eg for A , choice_list is [2, 1, 0] and for B , it is [5, 4, 3] . We get the offset for column through 3 * col_idx where 3 is the number of conditions.

Then we perform a select ion depending on these which gives what columns we should look for ( cols_in_df2 ). Rows to look for are the index of the series, so we choose the multipliers_in_df2 via loc with these.

Lastly we really multiply the series at hand with those multipliers and return.

This process happens for each column with apply .