Pandas中，如何对特定的连续列做一些算术计算

Question

In the following code, I like to calculate the total percentage change for Value only when Code is 'b'. The expected answer is 0.6 (which is 3/4 * 8/10).

import pandas as pd
import numpy as np
x = pd.DataFrame({'Code':['a', 'a', 'a', 'b', 'b', 'a', 'a', 'a', 'b', 'b', 'b', 'a', 'a'], 'Value': np.arange(13)})

   Code  Value
0     a      0
1     a      1
2     a      2
3     b      3
4     b      4
5     a      5
6     a      6
7     a      7
8     b      8
9     b      9
10    b     10
11    a     11
12    a     12

I tried with df.groupby , but as there are two groups of 'b', it does not do what I expected.

Thank you very much for your time in advance.

Answer 1

What you're trying to calculate requires you to group

1. consecutive rows
2. sharing a property about their column values.

Notice that grouping consecutive rows is grouping data based on a property of the index . A common and very flexible trick you can do in cases like this is to introduce a new column that stores the property of the index you care about.

In this case, you can track in a column how many times the value in the Code column has changed between consecutive rows:

(x.assign(code_changed=lambda df: df.Code != df.Code.shift(),
          ordered_code=lambda df: df.code_changed.cumsum())

    Code    Value   code_changed    ordered_code
0   a       0       True            1
1   a       1       False           1
2   a       2       False           1
3   b       3       True            2
4   b       4       False           2
5   a       5       True            3
6   a       6       False           3
7   a       7       False           3
8   b       8       True            4
9   b       9       False           4
10  b       10      False           4
11  a       11      True            5
12  a       12      False           5

The ordered_code column contains the exact grouping information you're looking for. You can then reach the output you're hoping for by restricting to rows with Code equal to 'b' and aggregating Value s:

(x.assign(code_changed=lambda df: df.Code != df.Code.shift(),
          ordered_code=lambda df: df.code_changed.cumsum())
  .pipe(lambda df: df[df.Code == 'b'])
  .groupby('ordered_code')
  .Value
  .agg(lambda values: values.iloc[0] / values.iloc[-1])
  .prod())

This outputs

0.6000000000000001

as desired.

Answer 2

Then take whatever your desired values are from pct_change and multiply them together as you wish.

pct_change = df.loc[df['Code'] == 'b'].pct_change()

Multiply the first and third value.

pct_change.iloc[[1]].values * pct_change.iloc[[3]].values                                                          

Or if you have multiple values you can write a loop to get different rows of pct_change.