如何计算 pandas 中行之间的条件百分比变化？

Question

Here's my dataframe:

df = pd.DataFrame({'Period': ['1_Baseline', '1_Baseline', '1_Baseline', '2_Acute', '2_Acute', '2_Acute', '3_Chronic', '3_Chronic', '3_Chronic', '4_Discontinuation', '4_Discontinuation', '4_Discontinuation'],
               'Subject': [1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3],
               'Amount': [24, 52, 34, 95, 98, 54, 32, 20, 16, 52, 34, 95]})

I want to create a column that contains a percent change in Amount within each Subject, for each Period, relative to Baseline. So, for Baseline, it would show how much the Amount changes for subject 1 from Baseline to Acute, and from 1_Baseline to 3_Chronic, and from 1_Baseline to 4_Discontinuation. It would do the same thing for each subject.

Here's what I tried:

df['pct_change'] = df.groupby(['Period'])['Amount'].pct_change()

But I get:

               Period  Subject  Amount  pct_change
0          1_Baseline        1      24         NaN
1          1_Baseline        2      52    1.166667
2          1_Baseline        3      34   -0.346154
3             2_Acute        1      95    1.794118
4             2_Acute        2      98    0.031579
5             2_Acute        3      54   -0.448980
6           3_Chronic        1      32   -0.407407
7           3_Chronic        2      20   -0.375000
8           3_Chronic        3      16   -0.200000
9   4_Discontinuation        1      52    2.250000
10  4_Discontinuation        2      34   -0.346154
11  4_Discontinuation        3      95    1.794118

The results are not calculated within each Period, and are not relative to each Subject's previous Amount.

Expect Output:

               Period  Subject  Amount  pct_change
0          1_Baseline        1      24         NaN
1          1_Baseline        2      52         NaN
2          1_Baseline        3      34         NaN
3             2_Acute        1      95         2.958333333
4             2_Acute        2      98         0.884615385
5             2_Acute        3      54         0.588235294
6           3_Chronic        1      32         0.333333333
7           3_Chronic        2      20        -0.615384615
8           3_Chronic        3      16        -0.529411765
9   4_Discontinuation        1      52         1.166666667
10  4_Discontinuation        2      34        -0.346153846
11  4_Discontinuation        3      95         1.794117647

Answer 1

IIUC, you want to divide Amount at every row with Subject==2 to Amount at Period==1_Baseline and Subject==2 . Here's my approach:

s = df.set_index(['Subject', 'Period']).Amount.unstack('Period')
df['pct_change'] = (s.div(s['1_Baseline'], axis='rows').sub(1)
                    .unstack().values
                   )

Output:

               Period  Subject  Amount  pct_change
0          1_Baseline        1      24    0.000000
1          1_Baseline        2      52    0.000000
2          1_Baseline        3      34    0.000000
3             2_Acute        1      95    2.958333
4             2_Acute        2      98    0.884615
5             2_Acute        3      54    0.588235
6           3_Chronic        1      32    0.333333
7           3_Chronic        2      20   -0.615385
8           3_Chronic        3      16   -0.529412
9   4_Discontinuation        1      52    1.166667
10  4_Discontinuation        2      34   -0.346154
11  4_Discontinuation        3      95    1.794118

Note that the order of the rows is very important. In this case, you do have the correct row order for this to work. If you are not certain about the order, then it's safer to merge:

s = df.set_index(['Subject', 'Period']).Amount.unstack('Period')
s = s.div(s['1_Baseline'], axis='rows').sub(1).unstack().reset_index(name='pct_change')

df.merge(s, on=['Period','Subject'], how='left')