42

I have a dataset with personal data such as name, height, weight and date of birth. I would build a graph with the number of people born in a particular month and year. I'm using python pandas to accomplish this and my strategy was to try to group by year and month and add using count. But the closest I got is to get the count of people by year or by month but not by both.

df['birthdate'].groupby(df.birthdate.dt.year).agg('count')

Other questions in stackoverflow point to a Grouper called TimeGrouper but searching in pandas documentation found nothing. Any idea?

Trenton McKinney
  • 56,955
  • 33
  • 144
  • 158
nsbm
  • 5,842
  • 6
  • 30
  • 45

5 Answers5

73

To group on multiple criteria, pass a list of the columns or criteria:

df['birthdate'].groupby([df.birthdate.dt.year, df.birthdate.dt.month]).agg('count')

Example:

In [165]:
df = pd.DataFrame({'birthdate':pd.date_range(start=dt.datetime(2015,12,20),end=dt.datetime(2016,3,1))})
df.groupby([df['birthdate'].dt.year, df['birthdate'].dt.month]).agg({'count'})

Out[165]:
                    birthdate
                        count
birthdate birthdate          
2015      12               12
2016      1                31
          2                29
          3                 1

UPDATE

As of version 0.23.0 the above code no longer works due to the restriction that multi-index level names must be unique, you now need to rename the levels in order for this to work:

In[107]:
df.groupby([df['birthdate'].dt.year.rename('year'), df['birthdate'].dt.month.rename('month')]).agg({'count'})

Out[107]: 
           birthdate
               count
year month          
2015 12           12
2016 1            31
     2            29
     3             1
EdChum
  • 376,765
  • 198
  • 813
  • 562
34

You can also use the "monthly" period with to_period with the dt accessor:

In [11]: df = pd.DataFrame({'birthdate': pd.date_range(start='20-12-2015', end='3-1-2016')})

In [12]: df['birthdate'].groupby(df.birthdate.dt.to_period("M")).agg('count')
Out[12]:
birthdate
2015-12    12
2016-01    31
2016-02    29
2016-03     1
Freq: M, Name: birthdate, dtype: int64

It's worth noting if the datetime is the index (rather than a column) you can use resample:

df.resample("M").count()
Andy Hayden
  • 359,921
  • 101
  • 625
  • 535
15

Another solution is to set birthdate as the index and resample:

import pandas as pd

df = pd.DataFrame({'birthdate': pd.date_range(start='20-12-2015', end='3-1-2016')})
df.set_index('birthdate').resample('MS').size()

Output:

birthdate
2015-12-01    12
2016-01-01    31
2016-02-01    29
2016-03-01     1
Freq: MS, dtype: int64
Alicia Garcia-Raboso
  • 13,193
  • 1
  • 43
  • 48
3

As of April 2019: This will work. Pandas version - 0.24.x

df.groupby([df.dates.dt.year.rename('year'), df.dates.dt.month.rename('month')]).size()

saran3h
  • 12,353
  • 4
  • 42
  • 54
0

Replace date and count fields with your respective column names. This piece of code will group, sum and sort based on the given parameters. You can also change the frequency to 1M or 2M and so on...

df[['date', 'count']].groupby(pd.Grouper(key='date', freq='1M')).sum().sort_values(by='date', ascending=True)['count']
user1775015
  • 179
  • 1
  • 6