sample pandas, keeping collected records together

Question

I'd like to (approximately) partition a large number of records in a dataframe. This is easily achieved using sample :

fraction = .7
df1     = df.sample(frac = fraction)
df2     = df.drop(df1.index)

Now for my question: Suppose I'd like to (randomly) partition the dataset but must also keep all records of a group together. It can be assumed that only a few records belong to each group, so that it will not interfere with the ability to randomly partition. An example, noting that the group only collects a few records and therefore does not impose significant constraint:

df = 
      group   value
0     'aaa'    48
1     'aaa'   -103
2     'aab'    20     
3     'aac'    21
4     'aac'    40
...
10000 'zzf'    220

Answer 1

Can you try groupby and then sample the groups as:

grouped = df.groupby('group') 
grouped.apply(lambda x: x.sample(frac=0.7)) –