如何按列分组并将组中的所有值复制到pandas中的一行？

Question

This is a sample of my dataset:

Consumer_num | billed_units  
29           | 984
29           | 1244
29           | 2323
29           | 1232
29           | 1150
30           | 3222
30           | 1444
30           | 2124

I want to group by consumer_num and then add all values (billed_units) of each group into new columns. So my required output:

Consumer_num | month 1 | month 2 | month 3 | month 4  | month 5  
29           | 984     | 1244     | 2323    | 1232     | 1150 
30           | 3222    | 1444     | 2124    | NaN      | NaN

This is what I've done so far:

group = df.groupby('consumer_num')['billed_units'].unique()
group[group.apply(lambda x: len(x)>1)]
df = group.to_frame()
print df

Output:

Consumer_num | billed_units  
29           | [984,1244,2323,1232,1150]
30           | [3222,1444,2124]

I don't know whether my approach is correct. If it's right, then I would like to know how I can separate billed_units of each consumer and then add to new columns as I've shown in my required output. Or is there a better method to achieve my required output?

Answer 1

solution

c = 'Consumer_num'
m = 'month {}'.format
df.set_index(
    [c, df.groupby(c).cumcount() + 1]
).billed_units.unstack().rename(columns=m).reset_index()

   Consumer_num  month 1  month 2  month 3  month 4  month 5
0            29    984.0   1244.0   2323.0   1232.0   1150.0
1            30   3222.0   1444.0   2124.0      NaN      NaN

how it works

• put 'Consumer_num' into a variable c for convenience
• put mapper function into a variable m for convenience
• setting index with two columns to make a pd.MultiIndex
  • I use groupby and cumcount to create a level to unstack with
  • then I unstack
• finally use the mapper function to rename the columns

---

response to comments

One approach for limiting the number of months is to use iloc . The following limits us to 3 months. You can adjust to take first 5. The nans should take care of themselves.

c = 'Consumer_num'
m = 'month {}'.format
df.set_index(
    [c, df.groupby(c).cumcount() + 1]
).billed_units.unstack().rename(columns=m).iloc[:, :3].reset_index()
#                                         ^..........^

   Consumer_num  month 1  month 2  month 3
0            29    984.0   1244.0   2323.0
1            30   3222.0   1444.0   2124.0

---

Or you could pre-process

c = 'Consumer_num'
m = 'month {}'.format
d1 = df.groupby(c).head(3)  # pre-process and take just first 3
d1.set_index(
    [c, d1.groupby(c).cumcount() + 1]
).billed_units.unstack().rename(columns=m).reset_index()

Answer 2

You could use pivot like

In [70]: dfm = df.assign(m=df.groupby('Consumer_num').cumcount().add(1))

In [71]: dfm.pivot('Consumer_num', 'm', 'billed_units').add_prefix('month ')
Out[71]:
m             month 1  month 2  month 3  month 4  month 5
Consumer_num
29              984.0   1244.0   2323.0   1232.0   1150.0
30             3222.0   1444.0   2124.0      NaN      NaN

---

Details

In [75]: df
Out[75]:
   Consumer_num  billed_units
0            29           984
1            29          1244
2            29          2323
3            29          1232
4            29          1150
5            30          3222
6            30          1444
7            30          2124

In [76]: dfm
Out[76]:
   Consumer_num  billed_units  m
0            29           984  1
1            29          1244  2
2            29          2323  3
3            29          1232  4
4            29          1150  5
5            30          3222  1
6            30          1444  2
7            30          2124  3